class GraphModule(torch.nn.Module):
    def forward(self, L_x_: "f32[1, 3, 1088, 1088][3551232, 1183744, 1088, 1]cuda:0", L_self_modules_patch_embed_modules_proj_parameters_weight_: "f32[128, 3, 4, 4][48, 16, 4, 1]cuda:0", L_self_modules_patch_embed_modules_proj_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_patch_embed_modules_norm_parameters_weight_: "f32[128][1]cuda:0", L_self_modules_patch_embed_modules_norm_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_weight_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_: "f32[384, 128][128, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_: "f32[384][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_: "f32[169, 4][4, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_: "f32[128, 128][128, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_weight_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_: "f32[512, 128][128, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_: "f32[128, 512][512, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_weight_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_: "f32[384, 128][128, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_: "f32[384][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_: "f32[169, 4][4, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_: "f32[128, 128][128, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_weight_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_: "f32[512, 128][128, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_: "f32[128, 512][512, 1]cuda:0", L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_0_modules_downsample_modules_reduction_parameters_weight_: "f32[256, 512][512, 1]cuda:0", L_self_modules_norm0_parameters_weight_: "f32[128][1]cuda:0", L_self_modules_norm0_parameters_bias_: "f32[128][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_weight_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_: "f32[768, 256][256, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_: "f32[768][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_: "f32[169, 8][8, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_: "f32[256, 256][256, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_weight_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_: "f32[1024, 256][256, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_: "f32[1024][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_: "f32[256, 1024][1024, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_weight_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_: "f32[768, 256][256, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_: "f32[768][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_: "f32[169, 8][8, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_: "f32[256, 256][256, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_weight_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_: "f32[1024, 256][256, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_: "f32[1024][1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_: "f32[256, 1024][1024, 1]cuda:0", L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_weight_: "f32[1024][1]cuda:0", L_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_bias_: "f32[1024][1]cuda:0", L_self_modules_layers_modules_1_modules_downsample_modules_reduction_parameters_weight_: "f32[512, 1024][1024, 1]cuda:0", L_self_modules_norm1_parameters_weight_: "f32[256][1]cuda:0", L_self_modules_norm1_parameters_bias_: "f32[256][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_weight_: "f32[1536, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_bias_: "f32[1536][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_parameters_relative_position_bias_table_: "f32[169, 16][16, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_buffers_relative_position_index_: "i64[49, 49][49, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_weight_: "f32[512, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_weight_: "f32[2048, 512][512, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_bias_: "f32[2048][1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_weight_: "f32[512, 2048][2048, 1]cuda:0", L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_bias_: "f32[512][1]cuda:0", L_self_modules_norm2_parameters_weight_: "f32[512][1]cuda:0", L_self_modules_norm2_parameters_bias_: "f32[512][1]cuda:0"):
        l_x_ = L_x_
        l_self_modules_patch_embed_modules_proj_parameters_weight_ = L_self_modules_patch_embed_modules_proj_parameters_weight_
        l_self_modules_patch_embed_modules_proj_parameters_bias_ = L_self_modules_patch_embed_modules_proj_parameters_bias_
        l_self_modules_patch_embed_modules_norm_parameters_weight_ = L_self_modules_patch_embed_modules_norm_parameters_weight_
        l_self_modules_patch_embed_modules_norm_parameters_bias_ = L_self_modules_patch_embed_modules_norm_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_weight_ = L_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_weight_
        l_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_bias_ = L_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_bias_
        l_self_modules_layers_modules_0_modules_downsample_modules_reduction_parameters_weight_ = L_self_modules_layers_modules_0_modules_downsample_modules_reduction_parameters_weight_
        l_self_modules_norm0_parameters_weight_ = L_self_modules_norm0_parameters_weight_
        l_self_modules_norm0_parameters_bias_ = L_self_modules_norm0_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_weight_ = L_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_weight_
        l_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_bias_ = L_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_bias_
        l_self_modules_layers_modules_1_modules_downsample_modules_reduction_parameters_weight_ = L_self_modules_layers_modules_1_modules_downsample_modules_reduction_parameters_weight_
        l_self_modules_norm1_parameters_weight_ = L_self_modules_norm1_parameters_weight_
        l_self_modules_norm1_parameters_bias_ = L_self_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_parameters_relative_position_bias_table_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_parameters_relative_position_bias_table_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_buffers_relative_position_index_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_buffers_relative_position_index_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_bias_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_weight_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_weight_
        l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_bias_ = L_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_bias_
        l_self_modules_norm2_parameters_weight_ = L_self_modules_norm2_parameters_weight_
        l_self_modules_norm2_parameters_bias_ = L_self_modules_norm2_parameters_bias_
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/conv.py:453 in _conv_forward, code: return F.conv2d(input, weight, bias, self.stride,
        x: "f16[1, 128, 272, 272][9469952, 73984, 272, 1]cuda:0" = torch.conv2d(l_x_, l_self_modules_patch_embed_modules_proj_parameters_weight_, l_self_modules_patch_embed_modules_proj_parameters_bias_, (4, 4), (0, 0), (1, 1), 1);  l_x_ = l_self_modules_patch_embed_modules_proj_parameters_weight_ = l_self_modules_patch_embed_modules_proj_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:520 in forward, code: x = x.flatten(2).transpose(1, 2)
        flatten: "f16[1, 128, 73984][9469952, 73984, 1]cuda:0" = x.flatten(2);  x = None
        x_1: "f16[1, 73984, 128][9469952, 1, 73984]cuda:0" = flatten.transpose(1, 2);  flatten = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_2: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.layer_norm(x_1, (128,), l_self_modules_patch_embed_modules_norm_parameters_weight_, l_self_modules_patch_embed_modules_norm_parameters_bias_, 1e-05);  x_1 = l_self_modules_patch_embed_modules_norm_parameters_weight_ = l_self_modules_patch_embed_modules_norm_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:522 in forward, code: x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)
        transpose_1: "f32[1, 128, 73984][9469952, 1, 128]cuda:0" = x_2.transpose(1, 2);  x_2 = None
        x_3: "f32[1, 128, 272, 272][9469952, 1, 34816, 128]cuda:0" = transpose_1.view(-1, 128, 272, 272);  transpose_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:703 in forward, code: x = x.flatten(2).transpose(1, 2)
        flatten_1: "f32[1, 128, 73984][9469952, 1, 128]cuda:0" = x_3.flatten(2);  x_3 = None
        x_4: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = flatten_1.transpose(1, 2);  flatten_1 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_5: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.dropout(x_4, 0.0, False, False);  x_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:435 in forward, code: H_tensor = torch.tensor(H, dtype=torch.float32, device=x.device)
        H_tensor: "f32[][]cuda:0" = torch.tensor(272, dtype = torch.float32, device = device(type='cuda', index=0))
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:436 in forward, code: W_tensor = torch.tensor(W, dtype=torch.float32, device=x.device)
        W_tensor: "f32[][]cuda:0" = torch.tensor(272, dtype = torch.float32, device = device(type='cuda', index=0))
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:437 in forward, code: Hp = torch.ceil(H_tensor / self.window_size) * self.window_size
        truediv: "f32[][]cuda:0" = H_tensor / 7;  H_tensor = None
        ceil: "f32[][]cuda:0" = torch.ceil(truediv);  truediv = None
        Hp: "f32[][]cuda:0" = ceil * 7;  ceil = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:438 in forward, code: Wp = torch.ceil(W_tensor / self.window_size) * self.window_size
        truediv_1: "f32[][]cuda:0" = W_tensor / 7;  W_tensor = None
        ceil_1: "f32[][]cuda:0" = torch.ceil(truediv_1);  truediv_1 = None
        Wp: "f32[][]cuda:0" = ceil_1 * 7;  ceil_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:440 in forward, code: Hp = Hp.to(torch.int32)  # Ensure Hp is an integer tensor
        Hp_1: "i32[][]cuda:0" = Hp.to(torch.int32);  Hp = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:441 in forward, code: Wp = Wp.to(torch.int32)  # Ensure Wp is an integer tensor
        Wp_1: "i32[][]cuda:0" = Wp.to(torch.int32);  Wp = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:443 in forward, code: img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
        img_mask: "f32[1, 273, 273, 1][74529, 273, 1, 1]cuda:0" = torch.zeros((1, Hp_1, Wp_1, 1), device = device(type='cuda', index=0));  Hp_1 = Wp_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:453 in forward, code: img_mask[:, h, w, :] = cnt
        img_mask[(slice(None, None, None), slice(0, -7, None), slice(0, -7, None), slice(None, None, None))] = 0;  setitem = img_mask
        img_mask[(slice(None, None, None), slice(0, -7, None), slice(-7, -3, None), slice(None, None, None))] = 1;  setitem_1 = img_mask
        img_mask[(slice(None, None, None), slice(0, -7, None), slice(-3, None, None), slice(None, None, None))] = 2;  setitem_2 = img_mask
        img_mask[(slice(None, None, None), slice(-7, -3, None), slice(0, -7, None), slice(None, None, None))] = 3;  setitem_3 = img_mask
        img_mask[(slice(None, None, None), slice(-7, -3, None), slice(-7, -3, None), slice(None, None, None))] = 4;  setitem_4 = img_mask
        img_mask[(slice(None, None, None), slice(-7, -3, None), slice(-3, None, None), slice(None, None, None))] = 5;  setitem_5 = img_mask
        img_mask[(slice(None, None, None), slice(-3, None, None), slice(0, -7, None), slice(None, None, None))] = 6;  setitem_6 = img_mask
        img_mask[(slice(None, None, None), slice(-3, None, None), slice(-7, -3, None), slice(None, None, None))] = 7;  setitem_7 = img_mask
        img_mask[(slice(None, None, None), slice(-3, None, None), slice(-3, None, None), slice(None, None, None))] = 8;  setitem_8 = img_mask
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_6: "f32[1, 39, 7, 39, 7, 1][74529, 1911, 273, 7, 1, 1]cuda:0" = img_mask.view(1, 39, 7, 39, 7, 1);  img_mask = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute: "f32[1, 39, 39, 7, 7, 1][74529, 1911, 7, 273, 1, 1]cuda:0" = x_6.permute(0, 1, 3, 2, 4, 5);  x_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous: "f32[1, 39, 39, 7, 7, 1][74529, 1911, 49, 7, 1, 1]cuda:0" = permute.contiguous();  permute = None
        windows: "f32[1521, 7, 7, 1][49, 7, 1, 1]cuda:0" = contiguous.view(-1, 7, 7, 1);  contiguous = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:458 in forward, code: mask_windows = mask_windows.view(-1,
        mask_windows: "f32[1521, 49][49, 1]cuda:0" = windows.view(-1, 49);  windows = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:460 in forward, code: attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
        unsqueeze: "f32[1521, 1, 49][49, 49, 1]cuda:0" = mask_windows.unsqueeze(1)
        unsqueeze_1: "f32[1521, 49, 1][49, 1, 1]cuda:0" = mask_windows.unsqueeze(2);  mask_windows = None
        attn_mask: "f32[1521, 49, 49][2401, 49, 1]cuda:0" = unsqueeze - unsqueeze_1;  unsqueeze = unsqueeze_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:461 in forward, code: attn_mask = attn_mask.masked_fill(attn_mask != 0,
        ne: "b8[1521, 49, 49][2401, 49, 1]cuda:0" = attn_mask != 0
        masked_fill: "f32[1521, 49, 49][2401, 49, 1]cuda:0" = attn_mask.masked_fill(ne, -100.0);  ne = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:463 in forward, code: attn_mask == 0, float(0.0))
        eq: "b8[1521, 49, 49][2401, 49, 1]cuda:0" = attn_mask == 0;  attn_mask = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:462 in forward, code: float(-100.0)).masked_fill(
        attn_mask_1: "f32[1521, 49, 49][2401, 49, 1]cuda:0" = masked_fill.masked_fill(eq, 0.0);  masked_fill = eq = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_7: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.layer_norm(x_5, (128,), l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_8: "f32[1, 272, 272, 128][9469952, 34816, 128, 1]cuda:0" = x_7.view(1, 272, 272, 128);  x_7 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_9: "f32[1, 273, 273, 128][9539712, 34944, 128, 1]cuda:0" = torch._C._nn.pad(x_8, (0, 0, 0, 1, 0, 1), 'constant', None);  x_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_10: "f32[1, 39, 7, 39, 7, 128][9539712, 244608, 34944, 896, 128, 1]cuda:0" = x_9.view(1, 39, 7, 39, 7, 128);  x_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_1: "f32[1, 39, 39, 7, 7, 128][9539712, 244608, 896, 34944, 128, 1]cuda:0" = x_10.permute(0, 1, 3, 2, 4, 5);  x_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_1: "f32[1, 39, 39, 7, 7, 128][9539712, 244608, 6272, 896, 128, 1]cuda:0" = permute_1.contiguous();  permute_1 = None
        windows_1: "f32[1521, 7, 7, 128][6272, 896, 128, 1]cuda:0" = contiguous_1.view(-1, 7, 7, 128);  contiguous_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows: "f32[1521, 49, 128][6272, 128, 1]cuda:0" = windows_1.view(-1, 49, 128);  windows_1 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear: "f16[1521, 49, 384][18816, 384, 1]cuda:0" = torch._C._nn.linear(x_windows, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_);  x_windows = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape: "f16[1521, 49, 3, 4, 32][18816, 384, 128, 32, 1]cuda:0" = linear.reshape(1521, 49, 3, 4, 32);  linear = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv: "f16[3, 1521, 4, 49, 32][128, 18816, 32, 384, 1]cuda:0" = reshape.permute(2, 0, 3, 1, 4);  reshape = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q: "f16[1521, 4, 49, 32][18816, 32, 384, 1]cuda:0" = qkv[0]
        k: "f16[1521, 4, 49, 32][18816, 32, 384, 1]cuda:0" = qkv[1]
        v: "f16[1521, 4, 49, 32][18816, 32, 384, 1]cuda:0" = qkv[2];  qkv = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_1: "f16[1521, 4, 49, 32][6272, 32, 128, 1]cuda:0" = q * 0.1767766952966369;  q = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_3: "f16[1521, 4, 32, 49][18816, 32, 1, 384]cuda:0" = k.transpose(-2, -1);  k = None
        attn: "f16[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = q_1 @ transpose_3;  q_1 = transpose_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_8: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_3: "f32[2401, 4][4, 1]cuda:0" = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_[view_8];  l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_ = view_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias: "f32[49, 49, 4][196, 4, 1]cuda:0" = getitem_3.view(49, 49, -1);  getitem_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_3: "f32[4, 49, 49][1, 196, 4]cuda:0" = relative_position_bias.permute(2, 0, 1);  relative_position_bias = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_1: "f32[4, 49, 49][2401, 49, 1]cuda:0" = permute_3.contiguous();  permute_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_2: "f32[1, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = relative_position_bias_1.unsqueeze(0);  relative_position_bias_1 = None
        attn_1: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = attn + unsqueeze_2;  attn = unsqueeze_2 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_2: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_1, -1, _stacklevel = 5);  attn_1 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_3: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_2, 0.0, False, False);  attn_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_1: "f16[1521, 4, 49, 32][6272, 1568, 32, 1]cuda:0" = attn_3 @ v;  attn_3 = v = None
        transpose_4: "f16[1521, 49, 4, 32][6272, 32, 1568, 1]cuda:0" = matmul_1.transpose(1, 2);  matmul_1 = None
        x_11: "f16[1521, 49, 128][6272, 128, 1]cuda:0" = transpose_4.reshape(1521, 49, 128);  transpose_4 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_12: "f16[1521, 49, 128][6272, 128, 1]cuda:0" = torch._C._nn.linear(x_11, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_);  x_11 = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_13: "f16[1521, 49, 128][6272, 128, 1]cuda:0" = torch.nn.functional.dropout(x_12, 0.0, False, False);  x_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows: "f16[1521, 7, 7, 128][6272, 896, 128, 1]cuda:0" = x_13.view(-1, 7, 7, 128);  x_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_14: "f16[1, 39, 39, 7, 7, 128][9539712, 244608, 6272, 896, 128, 1]cuda:0" = attn_windows.view(1, 39, 39, 7, 7, -1);  attn_windows = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_4: "f16[1, 39, 7, 39, 7, 128][9539712, 244608, 896, 6272, 128, 1]cuda:0" = x_14.permute(0, 1, 3, 2, 4, 5);  x_14 = None
        contiguous_3: "f16[1, 39, 7, 39, 7, 128][9539712, 244608, 34944, 896, 128, 1]cuda:0" = permute_4.contiguous();  permute_4 = None
        x_15: "f16[1, 273, 273, 128][9539712, 34944, 128, 1]cuda:0" = contiguous_3.view(1, 273, 273, -1);  contiguous_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_4: "f16[1, 272, 272, 128][9539712, 34944, 128, 1]cuda:0" = x_15[(slice(None, None, None), slice(None, 272, None), slice(None, 272, None), slice(None, None, None))];  x_15 = None
        x_16: "f16[1, 272, 272, 128][9469952, 34816, 128, 1]cuda:0" = getitem_4.contiguous();  getitem_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_17: "f16[1, 73984, 128][9469952, 128, 1]cuda:0" = x_16.view(1, 73984, 128);  x_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_18: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = x_5 + x_17;  x_5 = x_17 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_2: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.layer_norm(x_18, (128,), l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_19: "f16[1, 73984, 512][37879808, 512, 1]cuda:0" = torch._C._nn.linear(layer_norm_2, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_2 = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_20: "f16[1, 73984, 512][37879808, 512, 1]cuda:0" = torch._C._nn.gelu(x_19, approximate = 'none');  x_19 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_21: "f16[1, 73984, 512][37879808, 512, 1]cuda:0" = torch.nn.functional.dropout(x_20, 0.0, False, False);  x_20 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_22: "f16[1, 73984, 128][9469952, 128, 1]cuda:0" = torch._C._nn.linear(x_21, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_);  x_21 = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_23: "f16[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.dropout(x_22, 0.0, False, False);  x_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_24: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = x_18 + x_23;  x_18 = x_23 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_25: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.layer_norm(x_24, (128,), l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_26: "f32[1, 272, 272, 128][9469952, 34816, 128, 1]cuda:0" = x_25.view(1, 272, 272, 128);  x_25 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_27: "f32[1, 273, 273, 128][9539712, 34944, 128, 1]cuda:0" = torch._C._nn.pad(x_26, (0, 0, 0, 1, 0, 1), 'constant', None);  x_26 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x: "f32[1, 273, 273, 128][9539712, 34944, 128, 1]cuda:0" = torch.roll(x_27, shifts = (-3, -3), dims = (1, 2));  x_27 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_28: "f32[1, 39, 7, 39, 7, 128][9539712, 244608, 34944, 896, 128, 1]cuda:0" = shifted_x.view(1, 39, 7, 39, 7, 128);  shifted_x = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_5: "f32[1, 39, 39, 7, 7, 128][9539712, 244608, 896, 34944, 128, 1]cuda:0" = x_28.permute(0, 1, 3, 2, 4, 5);  x_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_5: "f32[1, 39, 39, 7, 7, 128][9539712, 244608, 6272, 896, 128, 1]cuda:0" = permute_5.contiguous();  permute_5 = None
        windows_2: "f32[1521, 7, 7, 128][6272, 896, 128, 1]cuda:0" = contiguous_5.view(-1, 7, 7, 128);  contiguous_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_1: "f32[1521, 49, 128][6272, 128, 1]cuda:0" = windows_2.view(-1, 49, 128);  windows_2 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_4: "f16[1521, 49, 384][18816, 384, 1]cuda:0" = torch._C._nn.linear(x_windows_1, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_);  x_windows_1 = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_2: "f16[1521, 49, 3, 4, 32][18816, 384, 128, 32, 1]cuda:0" = linear_4.reshape(1521, 49, 3, 4, 32);  linear_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_1: "f16[3, 1521, 4, 49, 32][128, 18816, 32, 384, 1]cuda:0" = reshape_2.permute(2, 0, 3, 1, 4);  reshape_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_2: "f16[1521, 4, 49, 32][18816, 32, 384, 1]cuda:0" = qkv_1[0]
        k_1: "f16[1521, 4, 49, 32][18816, 32, 384, 1]cuda:0" = qkv_1[1]
        v_1: "f16[1521, 4, 49, 32][18816, 32, 384, 1]cuda:0" = qkv_1[2];  qkv_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_3: "f16[1521, 4, 49, 32][6272, 32, 128, 1]cuda:0" = q_2 * 0.1767766952966369;  q_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_5: "f16[1521, 4, 32, 49][18816, 32, 1, 384]cuda:0" = k_1.transpose(-2, -1);  k_1 = None
        attn_4: "f16[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = q_3 @ transpose_5;  q_3 = transpose_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_18: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_8: "f32[2401, 4][4, 1]cuda:0" = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_[view_18];  l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_ = view_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_2: "f32[49, 49, 4][196, 4, 1]cuda:0" = getitem_8.view(49, 49, -1);  getitem_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_7: "f32[4, 49, 49][1, 196, 4]cuda:0" = relative_position_bias_2.permute(2, 0, 1);  relative_position_bias_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_3: "f32[4, 49, 49][2401, 49, 1]cuda:0" = permute_7.contiguous();  permute_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_3: "f32[1, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = relative_position_bias_3.unsqueeze(0);  relative_position_bias_3 = None
        attn_5: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = attn_4 + unsqueeze_3;  attn_4 = unsqueeze_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_20: "f32[1, 1521, 4, 49, 49][14607684, 9604, 2401, 49, 1]cuda:0" = attn_5.view(1, 1521, 4, 49, 49);  attn_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_4: "f32[1521, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_1.unsqueeze(1);  attn_mask_1 = None
        unsqueeze_5: "f32[1, 1521, 1, 49, 49][3651921, 2401, 2401, 49, 1]cuda:0" = unsqueeze_4.unsqueeze(0);  unsqueeze_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_6: "f32[1, 1521, 4, 49, 49][14607684, 9604, 2401, 49, 1]cuda:0" = view_20 + unsqueeze_5;  view_20 = unsqueeze_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_7: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = attn_6.view(-1, 4, 49, 49);  attn_6 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_8: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_7, -1, _stacklevel = 5);  attn_7 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_9: "f32[1521, 4, 49, 49][9604, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_8, 0.0, False, False);  attn_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_3: "f16[1521, 4, 49, 32][6272, 1568, 32, 1]cuda:0" = attn_9 @ v_1;  attn_9 = v_1 = None
        transpose_6: "f16[1521, 49, 4, 32][6272, 32, 1568, 1]cuda:0" = matmul_3.transpose(1, 2);  matmul_3 = None
        x_29: "f16[1521, 49, 128][6272, 128, 1]cuda:0" = transpose_6.reshape(1521, 49, 128);  transpose_6 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_30: "f16[1521, 49, 128][6272, 128, 1]cuda:0" = torch._C._nn.linear(x_29, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_);  x_29 = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_31: "f16[1521, 49, 128][6272, 128, 1]cuda:0" = torch.nn.functional.dropout(x_30, 0.0, False, False);  x_30 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_1: "f16[1521, 7, 7, 128][6272, 896, 128, 1]cuda:0" = x_31.view(-1, 7, 7, 128);  x_31 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_32: "f16[1, 39, 39, 7, 7, 128][9539712, 244608, 6272, 896, 128, 1]cuda:0" = attn_windows_1.view(1, 39, 39, 7, 7, -1);  attn_windows_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_8: "f16[1, 39, 7, 39, 7, 128][9539712, 244608, 896, 6272, 128, 1]cuda:0" = x_32.permute(0, 1, 3, 2, 4, 5);  x_32 = None
        contiguous_7: "f16[1, 39, 7, 39, 7, 128][9539712, 244608, 34944, 896, 128, 1]cuda:0" = permute_8.contiguous();  permute_8 = None
        x_33: "f16[1, 273, 273, 128][9539712, 34944, 128, 1]cuda:0" = contiguous_7.view(1, 273, 273, -1);  contiguous_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_34: "f16[1, 273, 273, 128][9539712, 34944, 128, 1]cuda:0" = torch.roll(x_33, shifts = (3, 3), dims = (1, 2));  x_33 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_9: "f16[1, 272, 272, 128][9539712, 34944, 128, 1]cuda:0" = x_34[(slice(None, None, None), slice(None, 272, None), slice(None, 272, None), slice(None, None, None))];  x_34 = None
        x_35: "f16[1, 272, 272, 128][9469952, 34816, 128, 1]cuda:0" = getitem_9.contiguous();  getitem_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_36: "f16[1, 73984, 128][9469952, 128, 1]cuda:0" = x_35.view(1, 73984, 128);  x_35 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_37: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = x_24 + x_36;  x_24 = x_36 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_4: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.layer_norm(x_37, (128,), l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_38: "f16[1, 73984, 512][37879808, 512, 1]cuda:0" = torch._C._nn.linear(layer_norm_4, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_4 = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_39: "f16[1, 73984, 512][37879808, 512, 1]cuda:0" = torch._C._nn.gelu(x_38, approximate = 'none');  x_38 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_40: "f16[1, 73984, 512][37879808, 512, 1]cuda:0" = torch.nn.functional.dropout(x_39, 0.0, False, False);  x_39 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_41: "f16[1, 73984, 128][9469952, 128, 1]cuda:0" = torch._C._nn.linear(x_40, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_);  x_40 = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_0_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_42: "f16[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.dropout(x_41, 0.0, False, False);  x_41 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_43: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = x_37 + x_42;  x_37 = x_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:344 in forward, code: x = x.view(B, H, W, C)
        x_44: "f32[1, 272, 272, 128][9469952, 34816, 128, 1]cuda:0" = x_43.view(1, 272, 272, 128)
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:351 in forward, code: x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
        x0: "f32[1, 136, 136, 128][9469952, 69632, 256, 1]cuda:0" = x_44[(slice(None, None, None), slice(0, None, 2), slice(0, None, 2), slice(None, None, None))]
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:352 in forward, code: x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
        x1: "f32[1, 136, 136, 128][9469952, 69632, 256, 1]cuda:0" = x_44[(slice(None, None, None), slice(1, None, 2), slice(0, None, 2), slice(None, None, None))]
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:353 in forward, code: x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
        x2: "f32[1, 136, 136, 128][9469952, 69632, 256, 1]cuda:0" = x_44[(slice(None, None, None), slice(0, None, 2), slice(1, None, 2), slice(None, None, None))]
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:354 in forward, code: x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
        x3: "f32[1, 136, 136, 128][9469952, 69632, 256, 1]cuda:0" = x_44[(slice(None, None, None), slice(1, None, 2), slice(1, None, 2), slice(None, None, None))];  x_44 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:355 in forward, code: x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
        x_45: "f32[1, 136, 136, 512][9469952, 69632, 512, 1]cuda:0" = torch.cat([x0, x1, x2, x3], -1);  x0 = x1 = x2 = x3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:356 in forward, code: x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
        x_46: "f32[1, 18496, 512][9469952, 512, 1]cuda:0" = x_45.view(1, -1, 512);  x_45 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_47: "f32[1, 18496, 512][9469952, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_46, (512,), l_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_weight_, l_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_bias_, 1e-05);  x_46 = l_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_weight_ = l_self_modules_layers_modules_0_modules_downsample_modules_norm_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_48: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = torch._C._nn.linear(x_47, l_self_modules_layers_modules_0_modules_downsample_modules_reduction_parameters_weight_, None);  x_47 = l_self_modules_layers_modules_0_modules_downsample_modules_reduction_parameters_weight_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_out: "f32[1, 73984, 128][9469952, 128, 1]cuda:0" = torch.nn.functional.layer_norm(x_43, (128,), l_self_modules_norm0_parameters_weight_, l_self_modules_norm0_parameters_bias_, 1e-05);  x_43 = l_self_modules_norm0_parameters_weight_ = l_self_modules_norm0_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:715 in forward, code: out = x_out.view(-1, H, W,
        view_28: "f32[1, 272, 272, 128][9469952, 34816, 128, 1]cuda:0" = x_out.view(-1, 272, 272, 128);  x_out = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:716 in forward, code: self.num_features[i]).permute(0, 3, 1,
        permute_9: "f32[1, 128, 272, 272][9469952, 1, 34816, 128]cuda:0" = view_28.permute(0, 3, 1, 2);  view_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:717 in forward, code: 2).contiguous()
        out: "f32[1, 128, 272, 272][9469952, 73984, 272, 1]cuda:0" = permute_9.contiguous();  permute_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:435 in forward, code: H_tensor = torch.tensor(H, dtype=torch.float32, device=x.device)
        H_tensor_1: "f32[][]cuda:0" = torch.tensor(136, dtype = torch.float32, device = device(type='cuda', index=0))
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:436 in forward, code: W_tensor = torch.tensor(W, dtype=torch.float32, device=x.device)
        W_tensor_1: "f32[][]cuda:0" = torch.tensor(136, dtype = torch.float32, device = device(type='cuda', index=0))
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:437 in forward, code: Hp = torch.ceil(H_tensor / self.window_size) * self.window_size
        truediv_2: "f32[][]cuda:0" = H_tensor_1 / 7;  H_tensor_1 = None
        ceil_2: "f32[][]cuda:0" = torch.ceil(truediv_2);  truediv_2 = None
        Hp_2: "f32[][]cuda:0" = ceil_2 * 7;  ceil_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:438 in forward, code: Wp = torch.ceil(W_tensor / self.window_size) * self.window_size
        truediv_3: "f32[][]cuda:0" = W_tensor_1 / 7;  W_tensor_1 = None
        ceil_3: "f32[][]cuda:0" = torch.ceil(truediv_3);  truediv_3 = None
        Wp_2: "f32[][]cuda:0" = ceil_3 * 7;  ceil_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:440 in forward, code: Hp = Hp.to(torch.int32)  # Ensure Hp is an integer tensor
        Hp_3: "i32[][]cuda:0" = Hp_2.to(torch.int32);  Hp_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:441 in forward, code: Wp = Wp.to(torch.int32)  # Ensure Wp is an integer tensor
        Wp_3: "i32[][]cuda:0" = Wp_2.to(torch.int32);  Wp_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:443 in forward, code: img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
        img_mask_1: "f32[1, 140, 140, 1][19600, 140, 1, 1]cuda:0" = torch.zeros((1, Hp_3, Wp_3, 1), device = device(type='cuda', index=0));  Hp_3 = Wp_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:453 in forward, code: img_mask[:, h, w, :] = cnt
        img_mask_1[(slice(None, None, None), slice(0, -7, None), slice(0, -7, None), slice(None, None, None))] = 0;  setitem_9 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(0, -7, None), slice(-7, -3, None), slice(None, None, None))] = 1;  setitem_10 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(0, -7, None), slice(-3, None, None), slice(None, None, None))] = 2;  setitem_11 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(-7, -3, None), slice(0, -7, None), slice(None, None, None))] = 3;  setitem_12 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(-7, -3, None), slice(-7, -3, None), slice(None, None, None))] = 4;  setitem_13 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(-7, -3, None), slice(-3, None, None), slice(None, None, None))] = 5;  setitem_14 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(-3, None, None), slice(0, -7, None), slice(None, None, None))] = 6;  setitem_15 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(-3, None, None), slice(-7, -3, None), slice(None, None, None))] = 7;  setitem_16 = img_mask_1
        img_mask_1[(slice(None, None, None), slice(-3, None, None), slice(-3, None, None), slice(None, None, None))] = 8;  setitem_17 = img_mask_1
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_49: "f32[1, 20, 7, 20, 7, 1][19600, 980, 140, 7, 1, 1]cuda:0" = img_mask_1.view(1, 20, 7, 20, 7, 1);  img_mask_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_10: "f32[1, 20, 20, 7, 7, 1][19600, 980, 7, 140, 1, 1]cuda:0" = x_49.permute(0, 1, 3, 2, 4, 5);  x_49 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_10: "f32[1, 20, 20, 7, 7, 1][19600, 980, 49, 7, 1, 1]cuda:0" = permute_10.contiguous();  permute_10 = None
        windows_3: "f32[400, 7, 7, 1][49, 7, 1, 1]cuda:0" = contiguous_10.view(-1, 7, 7, 1);  contiguous_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:458 in forward, code: mask_windows = mask_windows.view(-1,
        mask_windows_1: "f32[400, 49][49, 1]cuda:0" = windows_3.view(-1, 49);  windows_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:460 in forward, code: attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
        unsqueeze_6: "f32[400, 1, 49][49, 49, 1]cuda:0" = mask_windows_1.unsqueeze(1)
        unsqueeze_7: "f32[400, 49, 1][49, 1, 1]cuda:0" = mask_windows_1.unsqueeze(2);  mask_windows_1 = None
        attn_mask_2: "f32[400, 49, 49][2401, 49, 1]cuda:0" = unsqueeze_6 - unsqueeze_7;  unsqueeze_6 = unsqueeze_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:461 in forward, code: attn_mask = attn_mask.masked_fill(attn_mask != 0,
        ne_1: "b8[400, 49, 49][2401, 49, 1]cuda:0" = attn_mask_2 != 0
        masked_fill_2: "f32[400, 49, 49][2401, 49, 1]cuda:0" = attn_mask_2.masked_fill(ne_1, -100.0);  ne_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:463 in forward, code: attn_mask == 0, float(0.0))
        eq_1: "b8[400, 49, 49][2401, 49, 1]cuda:0" = attn_mask_2 == 0;  attn_mask_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:462 in forward, code: float(-100.0)).masked_fill(
        attn_mask_3: "f32[400, 49, 49][2401, 49, 1]cuda:0" = masked_fill_2.masked_fill(eq_1, 0.0);  masked_fill_2 = eq_1 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_50: "f32[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.layer_norm(x_48, (256,), l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_51: "f32[1, 136, 136, 256][4734976, 34816, 256, 1]cuda:0" = x_50.view(1, 136, 136, 256);  x_50 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_52: "f32[1, 140, 140, 256][5017600, 35840, 256, 1]cuda:0" = torch._C._nn.pad(x_51, (0, 0, 0, 4, 0, 4), 'constant', None);  x_51 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_53: "f32[1, 20, 7, 20, 7, 256][5017600, 250880, 35840, 1792, 256, 1]cuda:0" = x_52.view(1, 20, 7, 20, 7, 256);  x_52 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_11: "f32[1, 20, 20, 7, 7, 256][5017600, 250880, 1792, 35840, 256, 1]cuda:0" = x_53.permute(0, 1, 3, 2, 4, 5);  x_53 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_11: "f32[1, 20, 20, 7, 7, 256][5017600, 250880, 12544, 1792, 256, 1]cuda:0" = permute_11.contiguous();  permute_11 = None
        windows_4: "f32[400, 7, 7, 256][12544, 1792, 256, 1]cuda:0" = contiguous_11.view(-1, 7, 7, 256);  contiguous_11 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_2: "f32[400, 49, 256][12544, 256, 1]cuda:0" = windows_4.view(-1, 49, 256);  windows_4 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_9: "f16[400, 49, 768][37632, 768, 1]cuda:0" = torch._C._nn.linear(x_windows_2, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_);  x_windows_2 = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_4: "f16[400, 49, 3, 8, 32][37632, 768, 256, 32, 1]cuda:0" = linear_9.reshape(400, 49, 3, 8, 32);  linear_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_2: "f16[3, 400, 8, 49, 32][256, 37632, 32, 768, 1]cuda:0" = reshape_4.permute(2, 0, 3, 1, 4);  reshape_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_4: "f16[400, 8, 49, 32][37632, 32, 768, 1]cuda:0" = qkv_2[0]
        k_2: "f16[400, 8, 49, 32][37632, 32, 768, 1]cuda:0" = qkv_2[1]
        v_2: "f16[400, 8, 49, 32][37632, 32, 768, 1]cuda:0" = qkv_2[2];  qkv_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_5: "f16[400, 8, 49, 32][12544, 32, 256, 1]cuda:0" = q_4 * 0.1767766952966369;  q_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_7: "f16[400, 8, 32, 49][37632, 32, 1, 768]cuda:0" = k_2.transpose(-2, -1);  k_2 = None
        attn_10: "f16[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = q_5 @ transpose_7;  q_5 = transpose_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_36: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_17: "f32[2401, 8][8, 1]cuda:0" = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_[view_36];  l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_ = view_36 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_4: "f32[49, 49, 8][392, 8, 1]cuda:0" = getitem_17.view(49, 49, -1);  getitem_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_13: "f32[8, 49, 49][1, 392, 8]cuda:0" = relative_position_bias_4.permute(2, 0, 1);  relative_position_bias_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_5: "f32[8, 49, 49][2401, 49, 1]cuda:0" = permute_13.contiguous();  permute_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_8: "f32[1, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = relative_position_bias_5.unsqueeze(0);  relative_position_bias_5 = None
        attn_11: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = attn_10 + unsqueeze_8;  attn_10 = unsqueeze_8 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_12: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_11, -1, _stacklevel = 5);  attn_11 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_13: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_12, 0.0, False, False);  attn_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_5: "f16[400, 8, 49, 32][12544, 1568, 32, 1]cuda:0" = attn_13 @ v_2;  attn_13 = v_2 = None
        transpose_8: "f16[400, 49, 8, 32][12544, 32, 1568, 1]cuda:0" = matmul_5.transpose(1, 2);  matmul_5 = None
        x_54: "f16[400, 49, 256][12544, 256, 1]cuda:0" = transpose_8.reshape(400, 49, 256);  transpose_8 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_55: "f16[400, 49, 256][12544, 256, 1]cuda:0" = torch._C._nn.linear(x_54, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_);  x_54 = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_56: "f16[400, 49, 256][12544, 256, 1]cuda:0" = torch.nn.functional.dropout(x_55, 0.0, False, False);  x_55 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_2: "f16[400, 7, 7, 256][12544, 1792, 256, 1]cuda:0" = x_56.view(-1, 7, 7, 256);  x_56 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_57: "f16[1, 20, 20, 7, 7, 256][5017600, 250880, 12544, 1792, 256, 1]cuda:0" = attn_windows_2.view(1, 20, 20, 7, 7, -1);  attn_windows_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_14: "f16[1, 20, 7, 20, 7, 256][5017600, 250880, 1792, 12544, 256, 1]cuda:0" = x_57.permute(0, 1, 3, 2, 4, 5);  x_57 = None
        contiguous_13: "f16[1, 20, 7, 20, 7, 256][5017600, 250880, 35840, 1792, 256, 1]cuda:0" = permute_14.contiguous();  permute_14 = None
        x_58: "f16[1, 140, 140, 256][5017600, 35840, 256, 1]cuda:0" = contiguous_13.view(1, 140, 140, -1);  contiguous_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_18: "f16[1, 136, 136, 256][5017600, 35840, 256, 1]cuda:0" = x_58[(slice(None, None, None), slice(None, 136, None), slice(None, 136, None), slice(None, None, None))];  x_58 = None
        x_59: "f16[1, 136, 136, 256][4734976, 34816, 256, 1]cuda:0" = getitem_18.contiguous();  getitem_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_60: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = x_59.view(1, 18496, 256);  x_59 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_61: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = x_48 + x_60;  x_48 = x_60 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_8: "f32[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.layer_norm(x_61, (256,), l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_62: "f16[1, 18496, 1024][18939904, 1024, 1]cuda:0" = torch._C._nn.linear(layer_norm_8, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_8 = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_63: "f16[1, 18496, 1024][18939904, 1024, 1]cuda:0" = torch._C._nn.gelu(x_62, approximate = 'none');  x_62 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_64: "f16[1, 18496, 1024][18939904, 1024, 1]cuda:0" = torch.nn.functional.dropout(x_63, 0.0, False, False);  x_63 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_65: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = torch._C._nn.linear(x_64, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_);  x_64 = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_66: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.dropout(x_65, 0.0, False, False);  x_65 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_67: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = x_61 + x_66;  x_61 = x_66 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_68: "f32[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.layer_norm(x_67, (256,), l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_69: "f32[1, 136, 136, 256][4734976, 34816, 256, 1]cuda:0" = x_68.view(1, 136, 136, 256);  x_68 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_70: "f32[1, 140, 140, 256][5017600, 35840, 256, 1]cuda:0" = torch._C._nn.pad(x_69, (0, 0, 0, 4, 0, 4), 'constant', None);  x_69 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_1: "f32[1, 140, 140, 256][5017600, 35840, 256, 1]cuda:0" = torch.roll(x_70, shifts = (-3, -3), dims = (1, 2));  x_70 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_71: "f32[1, 20, 7, 20, 7, 256][5017600, 250880, 35840, 1792, 256, 1]cuda:0" = shifted_x_1.view(1, 20, 7, 20, 7, 256);  shifted_x_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_15: "f32[1, 20, 20, 7, 7, 256][5017600, 250880, 1792, 35840, 256, 1]cuda:0" = x_71.permute(0, 1, 3, 2, 4, 5);  x_71 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_15: "f32[1, 20, 20, 7, 7, 256][5017600, 250880, 12544, 1792, 256, 1]cuda:0" = permute_15.contiguous();  permute_15 = None
        windows_5: "f32[400, 7, 7, 256][12544, 1792, 256, 1]cuda:0" = contiguous_15.view(-1, 7, 7, 256);  contiguous_15 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_3: "f32[400, 49, 256][12544, 256, 1]cuda:0" = windows_5.view(-1, 49, 256);  windows_5 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_13: "f16[400, 49, 768][37632, 768, 1]cuda:0" = torch._C._nn.linear(x_windows_3, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_);  x_windows_3 = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_6: "f16[400, 49, 3, 8, 32][37632, 768, 256, 32, 1]cuda:0" = linear_13.reshape(400, 49, 3, 8, 32);  linear_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_3: "f16[3, 400, 8, 49, 32][256, 37632, 32, 768, 1]cuda:0" = reshape_6.permute(2, 0, 3, 1, 4);  reshape_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_6: "f16[400, 8, 49, 32][37632, 32, 768, 1]cuda:0" = qkv_3[0]
        k_3: "f16[400, 8, 49, 32][37632, 32, 768, 1]cuda:0" = qkv_3[1]
        v_3: "f16[400, 8, 49, 32][37632, 32, 768, 1]cuda:0" = qkv_3[2];  qkv_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_7: "f16[400, 8, 49, 32][12544, 32, 256, 1]cuda:0" = q_6 * 0.1767766952966369;  q_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_9: "f16[400, 8, 32, 49][37632, 32, 1, 768]cuda:0" = k_3.transpose(-2, -1);  k_3 = None
        attn_14: "f16[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = q_7 @ transpose_9;  q_7 = transpose_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_46: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_22: "f32[2401, 8][8, 1]cuda:0" = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_[view_46];  l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_ = view_46 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_6: "f32[49, 49, 8][392, 8, 1]cuda:0" = getitem_22.view(49, 49, -1);  getitem_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_17: "f32[8, 49, 49][1, 392, 8]cuda:0" = relative_position_bias_6.permute(2, 0, 1);  relative_position_bias_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_7: "f32[8, 49, 49][2401, 49, 1]cuda:0" = permute_17.contiguous();  permute_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_9: "f32[1, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = relative_position_bias_7.unsqueeze(0);  relative_position_bias_7 = None
        attn_15: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = attn_14 + unsqueeze_9;  attn_14 = unsqueeze_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_48: "f32[1, 400, 8, 49, 49][7683200, 19208, 2401, 49, 1]cuda:0" = attn_15.view(1, 400, 8, 49, 49);  attn_15 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_10: "f32[400, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_3.unsqueeze(1);  attn_mask_3 = None
        unsqueeze_11: "f32[1, 400, 1, 49, 49][960400, 2401, 2401, 49, 1]cuda:0" = unsqueeze_10.unsqueeze(0);  unsqueeze_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_16: "f32[1, 400, 8, 49, 49][7683200, 19208, 2401, 49, 1]cuda:0" = view_48 + unsqueeze_11;  view_48 = unsqueeze_11 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_17: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = attn_16.view(-1, 8, 49, 49);  attn_16 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_18: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_17, -1, _stacklevel = 5);  attn_17 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_19: "f32[400, 8, 49, 49][19208, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_18, 0.0, False, False);  attn_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_7: "f16[400, 8, 49, 32][12544, 1568, 32, 1]cuda:0" = attn_19 @ v_3;  attn_19 = v_3 = None
        transpose_10: "f16[400, 49, 8, 32][12544, 32, 1568, 1]cuda:0" = matmul_7.transpose(1, 2);  matmul_7 = None
        x_72: "f16[400, 49, 256][12544, 256, 1]cuda:0" = transpose_10.reshape(400, 49, 256);  transpose_10 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_73: "f16[400, 49, 256][12544, 256, 1]cuda:0" = torch._C._nn.linear(x_72, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_);  x_72 = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_74: "f16[400, 49, 256][12544, 256, 1]cuda:0" = torch.nn.functional.dropout(x_73, 0.0, False, False);  x_73 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_3: "f16[400, 7, 7, 256][12544, 1792, 256, 1]cuda:0" = x_74.view(-1, 7, 7, 256);  x_74 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_75: "f16[1, 20, 20, 7, 7, 256][5017600, 250880, 12544, 1792, 256, 1]cuda:0" = attn_windows_3.view(1, 20, 20, 7, 7, -1);  attn_windows_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_18: "f16[1, 20, 7, 20, 7, 256][5017600, 250880, 1792, 12544, 256, 1]cuda:0" = x_75.permute(0, 1, 3, 2, 4, 5);  x_75 = None
        contiguous_17: "f16[1, 20, 7, 20, 7, 256][5017600, 250880, 35840, 1792, 256, 1]cuda:0" = permute_18.contiguous();  permute_18 = None
        x_76: "f16[1, 140, 140, 256][5017600, 35840, 256, 1]cuda:0" = contiguous_17.view(1, 140, 140, -1);  contiguous_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_77: "f16[1, 140, 140, 256][5017600, 35840, 256, 1]cuda:0" = torch.roll(x_76, shifts = (3, 3), dims = (1, 2));  x_76 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_23: "f16[1, 136, 136, 256][5017600, 35840, 256, 1]cuda:0" = x_77[(slice(None, None, None), slice(None, 136, None), slice(None, 136, None), slice(None, None, None))];  x_77 = None
        x_78: "f16[1, 136, 136, 256][4734976, 34816, 256, 1]cuda:0" = getitem_23.contiguous();  getitem_23 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_79: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = x_78.view(1, 18496, 256);  x_78 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_80: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = x_67 + x_79;  x_67 = x_79 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_10: "f32[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.layer_norm(x_80, (256,), l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_81: "f16[1, 18496, 1024][18939904, 1024, 1]cuda:0" = torch._C._nn.linear(layer_norm_10, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_10 = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_82: "f16[1, 18496, 1024][18939904, 1024, 1]cuda:0" = torch._C._nn.gelu(x_81, approximate = 'none');  x_81 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_83: "f16[1, 18496, 1024][18939904, 1024, 1]cuda:0" = torch.nn.functional.dropout(x_82, 0.0, False, False);  x_82 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_84: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = torch._C._nn.linear(x_83, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_);  x_83 = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_1_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_85: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.dropout(x_84, 0.0, False, False);  x_84 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_86: "f16[1, 18496, 256][4734976, 256, 1]cuda:0" = x_80 + x_85;  x_80 = x_85 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:344 in forward, code: x = x.view(B, H, W, C)
        x_87: "f16[1, 136, 136, 256][4734976, 34816, 256, 1]cuda:0" = x_86.view(1, 136, 136, 256)
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:351 in forward, code: x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
        x0_1: "f16[1, 68, 68, 256][4734976, 69632, 512, 1]cuda:0" = x_87[(slice(None, None, None), slice(0, None, 2), slice(0, None, 2), slice(None, None, None))]
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:352 in forward, code: x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
        x1_1: "f16[1, 68, 68, 256][4734976, 69632, 512, 1]cuda:0" = x_87[(slice(None, None, None), slice(1, None, 2), slice(0, None, 2), slice(None, None, None))]
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:353 in forward, code: x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
        x2_1: "f16[1, 68, 68, 256][4734976, 69632, 512, 1]cuda:0" = x_87[(slice(None, None, None), slice(0, None, 2), slice(1, None, 2), slice(None, None, None))]
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:354 in forward, code: x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
        x3_1: "f16[1, 68, 68, 256][4734976, 69632, 512, 1]cuda:0" = x_87[(slice(None, None, None), slice(1, None, 2), slice(1, None, 2), slice(None, None, None))];  x_87 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:355 in forward, code: x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
        x_88: "f16[1, 68, 68, 1024][4734976, 69632, 1024, 1]cuda:0" = torch.cat([x0_1, x1_1, x2_1, x3_1], -1);  x0_1 = x1_1 = x2_1 = x3_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:356 in forward, code: x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
        x_89: "f16[1, 4624, 1024][4734976, 1024, 1]cuda:0" = x_88.view(1, -1, 1024);  x_88 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_90: "f32[1, 4624, 1024][4734976, 1024, 1]cuda:0" = torch.nn.functional.layer_norm(x_89, (1024,), l_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_weight_, l_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_bias_, 1e-05);  x_89 = l_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_weight_ = l_self_modules_layers_modules_1_modules_downsample_modules_norm_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_91: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_90, l_self_modules_layers_modules_1_modules_downsample_modules_reduction_parameters_weight_, None);  x_90 = l_self_modules_layers_modules_1_modules_downsample_modules_reduction_parameters_weight_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_out_1: "f32[1, 18496, 256][4734976, 256, 1]cuda:0" = torch.nn.functional.layer_norm(x_86, (256,), l_self_modules_norm1_parameters_weight_, l_self_modules_norm1_parameters_bias_, 1e-05);  x_86 = l_self_modules_norm1_parameters_weight_ = l_self_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:715 in forward, code: out = x_out.view(-1, H, W,
        view_56: "f32[1, 136, 136, 256][4734976, 34816, 256, 1]cuda:0" = x_out_1.view(-1, 136, 136, 256);  x_out_1 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:716 in forward, code: self.num_features[i]).permute(0, 3, 1,
        permute_19: "f32[1, 256, 136, 136][4734976, 1, 34816, 256]cuda:0" = view_56.permute(0, 3, 1, 2);  view_56 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:717 in forward, code: 2).contiguous()
        out_1: "f32[1, 256, 136, 136][4734976, 18496, 136, 1]cuda:0" = permute_19.contiguous();  permute_19 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:435 in forward, code: H_tensor = torch.tensor(H, dtype=torch.float32, device=x.device)
        H_tensor_2: "f32[][]cuda:0" = torch.tensor(68, dtype = torch.float32, device = device(type='cuda', index=0))
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:436 in forward, code: W_tensor = torch.tensor(W, dtype=torch.float32, device=x.device)
        W_tensor_2: "f32[][]cuda:0" = torch.tensor(68, dtype = torch.float32, device = device(type='cuda', index=0))
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:437 in forward, code: Hp = torch.ceil(H_tensor / self.window_size) * self.window_size
        truediv_4: "f32[][]cuda:0" = H_tensor_2 / 7;  H_tensor_2 = None
        ceil_4: "f32[][]cuda:0" = torch.ceil(truediv_4);  truediv_4 = None
        Hp_4: "f32[][]cuda:0" = ceil_4 * 7;  ceil_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:438 in forward, code: Wp = torch.ceil(W_tensor / self.window_size) * self.window_size
        truediv_5: "f32[][]cuda:0" = W_tensor_2 / 7;  W_tensor_2 = None
        ceil_5: "f32[][]cuda:0" = torch.ceil(truediv_5);  truediv_5 = None
        Wp_4: "f32[][]cuda:0" = ceil_5 * 7;  ceil_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:440 in forward, code: Hp = Hp.to(torch.int32)  # Ensure Hp is an integer tensor
        Hp_5: "i32[][]cuda:0" = Hp_4.to(torch.int32);  Hp_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:441 in forward, code: Wp = Wp.to(torch.int32)  # Ensure Wp is an integer tensor
        Wp_5: "i32[][]cuda:0" = Wp_4.to(torch.int32);  Wp_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:443 in forward, code: img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
        img_mask_2: "f32[1, 70, 70, 1][4900, 70, 1, 1]cuda:0" = torch.zeros((1, Hp_5, Wp_5, 1), device = device(type='cuda', index=0));  Hp_5 = Wp_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:453 in forward, code: img_mask[:, h, w, :] = cnt
        img_mask_2[(slice(None, None, None), slice(0, -7, None), slice(0, -7, None), slice(None, None, None))] = 0;  setitem_18 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(0, -7, None), slice(-7, -3, None), slice(None, None, None))] = 1;  setitem_19 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(0, -7, None), slice(-3, None, None), slice(None, None, None))] = 2;  setitem_20 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(-7, -3, None), slice(0, -7, None), slice(None, None, None))] = 3;  setitem_21 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(-7, -3, None), slice(-7, -3, None), slice(None, None, None))] = 4;  setitem_22 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(-7, -3, None), slice(-3, None, None), slice(None, None, None))] = 5;  setitem_23 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(-3, None, None), slice(0, -7, None), slice(None, None, None))] = 6;  setitem_24 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(-3, None, None), slice(-7, -3, None), slice(None, None, None))] = 7;  setitem_25 = img_mask_2
        img_mask_2[(slice(None, None, None), slice(-3, None, None), slice(-3, None, None), slice(None, None, None))] = 8;  setitem_26 = img_mask_2
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_92: "f32[1, 10, 7, 10, 7, 1][4900, 490, 70, 7, 1, 1]cuda:0" = img_mask_2.view(1, 10, 7, 10, 7, 1);  img_mask_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_20: "f32[1, 10, 10, 7, 7, 1][4900, 490, 7, 70, 1, 1]cuda:0" = x_92.permute(0, 1, 3, 2, 4, 5);  x_92 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_20: "f32[1, 10, 10, 7, 7, 1][4900, 490, 49, 7, 1, 1]cuda:0" = permute_20.contiguous();  permute_20 = None
        windows_6: "f32[100, 7, 7, 1][49, 7, 1, 1]cuda:0" = contiguous_20.view(-1, 7, 7, 1);  contiguous_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:458 in forward, code: mask_windows = mask_windows.view(-1,
        mask_windows_2: "f32[100, 49][49, 1]cuda:0" = windows_6.view(-1, 49);  windows_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:460 in forward, code: attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
        unsqueeze_12: "f32[100, 1, 49][49, 49, 1]cuda:0" = mask_windows_2.unsqueeze(1)
        unsqueeze_13: "f32[100, 49, 1][49, 1, 1]cuda:0" = mask_windows_2.unsqueeze(2);  mask_windows_2 = None
        attn_mask_4: "f32[100, 49, 49][2401, 49, 1]cuda:0" = unsqueeze_12 - unsqueeze_13;  unsqueeze_12 = unsqueeze_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:461 in forward, code: attn_mask = attn_mask.masked_fill(attn_mask != 0,
        ne_2: "b8[100, 49, 49][2401, 49, 1]cuda:0" = attn_mask_4 != 0
        masked_fill_4: "f32[100, 49, 49][2401, 49, 1]cuda:0" = attn_mask_4.masked_fill(ne_2, -100.0);  ne_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:463 in forward, code: attn_mask == 0, float(0.0))
        eq_2: "b8[100, 49, 49][2401, 49, 1]cuda:0" = attn_mask_4 == 0;  attn_mask_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:462 in forward, code: float(-100.0)).masked_fill(
        attn_mask_5: "f32[100, 49, 49][2401, 49, 1]cuda:0" = masked_fill_4.masked_fill(eq_2, 0.0);  masked_fill_4 = eq_2 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_93: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_91, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_94: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_93.view(1, 68, 68, 512);  x_93 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_95: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_94, (0, 0, 0, 2, 0, 2), 'constant', None);  x_94 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_96: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_95.view(1, 10, 7, 10, 7, 512);  x_95 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_21: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_96.permute(0, 1, 3, 2, 4, 5);  x_96 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_21: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_21.contiguous();  permute_21 = None
        windows_7: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_21.view(-1, 7, 7, 512);  contiguous_21 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_4: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_7.view(-1, 49, 512);  windows_7 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_18: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_4, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_);  x_windows_4 = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_8: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_18.reshape(100, 49, 3, 16, 32);  linear_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_4: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_8.permute(2, 0, 3, 1, 4);  reshape_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_8: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_4[0]
        k_4: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_4[1]
        v_4: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_4[2];  qkv_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_9: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_8 * 0.1767766952966369;  q_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_11: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_4.transpose(-2, -1);  k_4 = None
        attn_20: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_9 @ transpose_11;  q_9 = transpose_11 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_64: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_31: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_[view_64];  l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_parameters_relative_position_bias_table_ = view_64 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_8: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_31.view(49, 49, -1);  getitem_31 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_23: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_8.permute(2, 0, 1);  relative_position_bias_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_9: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_23.contiguous();  permute_23 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_14: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_9.unsqueeze(0);  relative_position_bias_9 = None
        attn_21: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_20 + unsqueeze_14;  attn_20 = unsqueeze_14 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_22: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_21, -1, _stacklevel = 5);  attn_21 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_23: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_22, 0.0, False, False);  attn_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_9: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_23 @ v_4;  attn_23 = v_4 = None
        transpose_12: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_9.transpose(1, 2);  matmul_9 = None
        x_97: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_12.reshape(100, 49, 512);  transpose_12 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_98: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_97, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_);  x_97 = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_99: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_98, 0.0, False, False);  x_98 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_4: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_99.view(-1, 7, 7, 512);  x_99 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_100: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_4.view(1, 10, 10, 7, 7, -1);  attn_windows_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_24: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_100.permute(0, 1, 3, 2, 4, 5);  x_100 = None
        contiguous_23: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_24.contiguous();  permute_24 = None
        x_101: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_23.view(1, 70, 70, -1);  contiguous_23 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_32: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_101[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_101 = None
        x_102: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_32.contiguous();  getitem_32 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_103: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_102.view(1, 4624, 512);  x_102 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_104: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_91 + x_103;  x_91 = x_103 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_14: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_104, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_105: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_14, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_14 = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_106: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_105, approximate = 'none');  x_105 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_107: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_106, 0.0, False, False);  x_106 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_108: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_107, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_);  x_107 = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_0_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_109: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_108, 0.0, False, False);  x_108 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_110: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_104 + x_109;  x_104 = x_109 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_111: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_110, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_112: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_111.view(1, 68, 68, 512);  x_111 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_113: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_112, (0, 0, 0, 2, 0, 2), 'constant', None);  x_112 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_2: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_113, shifts = (-3, -3), dims = (1, 2));  x_113 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_114: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_2.view(1, 10, 7, 10, 7, 512);  shifted_x_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_25: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_114.permute(0, 1, 3, 2, 4, 5);  x_114 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_25: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_25.contiguous();  permute_25 = None
        windows_8: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_25.view(-1, 7, 7, 512);  contiguous_25 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_5: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_8.view(-1, 49, 512);  windows_8 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_22: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_5, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_);  x_windows_5 = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_10: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_22.reshape(100, 49, 3, 16, 32);  linear_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_5: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_10.permute(2, 0, 3, 1, 4);  reshape_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_10: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_5[0]
        k_5: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_5[1]
        v_5: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_5[2];  qkv_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_11: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_10 * 0.1767766952966369;  q_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_13: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_5.transpose(-2, -1);  k_5 = None
        attn_24: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_11 @ transpose_13;  q_11 = transpose_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_74: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_36: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_[view_74];  l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_parameters_relative_position_bias_table_ = view_74 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_10: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_36.view(49, 49, -1);  getitem_36 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_27: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_10.permute(2, 0, 1);  relative_position_bias_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_11: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_27.contiguous();  permute_27 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_15: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_11.unsqueeze(0);  relative_position_bias_11 = None
        attn_25: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_24 + unsqueeze_15;  attn_24 = unsqueeze_15 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_76: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_25.view(1, 100, 16, 49, 49);  attn_25 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_16: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_17: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_16.unsqueeze(0);  unsqueeze_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_26: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_76 + unsqueeze_17;  view_76 = unsqueeze_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_27: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_26.view(-1, 16, 49, 49);  attn_26 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_28: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_27, -1, _stacklevel = 5);  attn_27 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_29: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_28, 0.0, False, False);  attn_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_11: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_29 @ v_5;  attn_29 = v_5 = None
        transpose_14: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_11.transpose(1, 2);  matmul_11 = None
        x_115: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_14.reshape(100, 49, 512);  transpose_14 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_116: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_115, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_);  x_115 = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_117: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_116, 0.0, False, False);  x_116 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_5: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_117.view(-1, 7, 7, 512);  x_117 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_118: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_5.view(1, 10, 10, 7, 7, -1);  attn_windows_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_28: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_118.permute(0, 1, 3, 2, 4, 5);  x_118 = None
        contiguous_27: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_28.contiguous();  permute_28 = None
        x_119: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_27.view(1, 70, 70, -1);  contiguous_27 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_120: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_119, shifts = (3, 3), dims = (1, 2));  x_119 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_37: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_120[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_120 = None
        x_121: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_37.contiguous();  getitem_37 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_122: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_121.view(1, 4624, 512);  x_121 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_123: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_110 + x_122;  x_110 = x_122 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_16: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_123, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_124: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_16, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_16 = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_125: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_124, approximate = 'none');  x_124 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_126: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_125, 0.0, False, False);  x_125 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_127: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_126, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_);  x_126 = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_1_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_128: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_127, 0.0, False, False);  x_127 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_129: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_123 + x_128;  x_123 = x_128 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_130: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_129, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_131: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_130.view(1, 68, 68, 512);  x_130 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_132: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_131, (0, 0, 0, 2, 0, 2), 'constant', None);  x_131 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_133: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_132.view(1, 10, 7, 10, 7, 512);  x_132 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_29: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_133.permute(0, 1, 3, 2, 4, 5);  x_133 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_29: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_29.contiguous();  permute_29 = None
        windows_9: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_29.view(-1, 7, 7, 512);  contiguous_29 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_6: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_9.view(-1, 49, 512);  windows_9 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_26: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_6, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_bias_);  x_windows_6 = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_12: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_26.reshape(100, 49, 3, 16, 32);  linear_26 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_6: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_12.permute(2, 0, 3, 1, 4);  reshape_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_12: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_6[0]
        k_6: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_6[1]
        v_6: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_6[2];  qkv_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_13: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_12 * 0.1767766952966369;  q_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_15: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_6.transpose(-2, -1);  k_6 = None
        attn_30: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_13 @ transpose_15;  q_13 = transpose_15 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_86: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_41: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_parameters_relative_position_bias_table_[view_86];  l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_parameters_relative_position_bias_table_ = view_86 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_12: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_41.view(49, 49, -1);  getitem_41 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_31: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_12.permute(2, 0, 1);  relative_position_bias_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_13: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_31.contiguous();  permute_31 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_18: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_13.unsqueeze(0);  relative_position_bias_13 = None
        attn_31: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_30 + unsqueeze_18;  attn_30 = unsqueeze_18 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_32: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_31, -1, _stacklevel = 5);  attn_31 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_33: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_32, 0.0, False, False);  attn_32 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_13: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_33 @ v_6;  attn_33 = v_6 = None
        transpose_16: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_13.transpose(1, 2);  matmul_13 = None
        x_134: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_16.reshape(100, 49, 512);  transpose_16 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_135: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_134, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_bias_);  x_134 = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_136: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_135, 0.0, False, False);  x_135 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_6: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_136.view(-1, 7, 7, 512);  x_136 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_137: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_6.view(1, 10, 10, 7, 7, -1);  attn_windows_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_32: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_137.permute(0, 1, 3, 2, 4, 5);  x_137 = None
        contiguous_31: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_32.contiguous();  permute_32 = None
        x_138: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_31.view(1, 70, 70, -1);  contiguous_31 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_42: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_138[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_138 = None
        x_139: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_42.contiguous();  getitem_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_140: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_139.view(1, 4624, 512);  x_139 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_141: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_129 + x_140;  x_129 = x_140 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_18: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_141, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_142: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_18, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_18 = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_143: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_142, approximate = 'none');  x_142 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_144: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_143, 0.0, False, False);  x_143 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_145: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_144, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_bias_);  x_144 = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_2_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_146: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_145, 0.0, False, False);  x_145 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_147: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_141 + x_146;  x_141 = x_146 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_148: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_147, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_149: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_148.view(1, 68, 68, 512);  x_148 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_150: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_149, (0, 0, 0, 2, 0, 2), 'constant', None);  x_149 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_3: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_150, shifts = (-3, -3), dims = (1, 2));  x_150 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_151: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_3.view(1, 10, 7, 10, 7, 512);  shifted_x_3 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_33: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_151.permute(0, 1, 3, 2, 4, 5);  x_151 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_33: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_33.contiguous();  permute_33 = None
        windows_10: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_33.view(-1, 7, 7, 512);  contiguous_33 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_7: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_10.view(-1, 49, 512);  windows_10 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_30: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_7, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_bias_);  x_windows_7 = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_14: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_30.reshape(100, 49, 3, 16, 32);  linear_30 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_7: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_14.permute(2, 0, 3, 1, 4);  reshape_14 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_14: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_7[0]
        k_7: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_7[1]
        v_7: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_7[2];  qkv_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_15: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_14 * 0.1767766952966369;  q_14 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_17: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_7.transpose(-2, -1);  k_7 = None
        attn_34: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_15 @ transpose_17;  q_15 = transpose_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_96: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_46: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_parameters_relative_position_bias_table_[view_96];  l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_parameters_relative_position_bias_table_ = view_96 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_14: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_46.view(49, 49, -1);  getitem_46 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_35: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_14.permute(2, 0, 1);  relative_position_bias_14 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_15: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_35.contiguous();  permute_35 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_19: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_15.unsqueeze(0);  relative_position_bias_15 = None
        attn_35: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_34 + unsqueeze_19;  attn_34 = unsqueeze_19 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_98: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_35.view(1, 100, 16, 49, 49);  attn_35 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_20: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_21: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_20.unsqueeze(0);  unsqueeze_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_36: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_98 + unsqueeze_21;  view_98 = unsqueeze_21 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_37: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_36.view(-1, 16, 49, 49);  attn_36 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_38: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_37, -1, _stacklevel = 5);  attn_37 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_39: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_38, 0.0, False, False);  attn_38 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_15: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_39 @ v_7;  attn_39 = v_7 = None
        transpose_18: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_15.transpose(1, 2);  matmul_15 = None
        x_152: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_18.reshape(100, 49, 512);  transpose_18 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_153: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_152, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_bias_);  x_152 = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_154: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_153, 0.0, False, False);  x_153 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_7: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_154.view(-1, 7, 7, 512);  x_154 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_155: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_7.view(1, 10, 10, 7, 7, -1);  attn_windows_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_36: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_155.permute(0, 1, 3, 2, 4, 5);  x_155 = None
        contiguous_35: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_36.contiguous();  permute_36 = None
        x_156: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_35.view(1, 70, 70, -1);  contiguous_35 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_157: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_156, shifts = (3, 3), dims = (1, 2));  x_156 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_47: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_157[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_157 = None
        x_158: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_47.contiguous();  getitem_47 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_159: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_158.view(1, 4624, 512);  x_158 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_160: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_147 + x_159;  x_147 = x_159 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_20: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_160, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_161: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_20, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_20 = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_162: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_161, approximate = 'none');  x_161 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_163: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_162, 0.0, False, False);  x_162 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_164: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_163, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_bias_);  x_163 = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_3_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_165: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_164, 0.0, False, False);  x_164 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_166: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_160 + x_165;  x_160 = x_165 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_167: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_166, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_168: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_167.view(1, 68, 68, 512);  x_167 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_169: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_168, (0, 0, 0, 2, 0, 2), 'constant', None);  x_168 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_170: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_169.view(1, 10, 7, 10, 7, 512);  x_169 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_37: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_170.permute(0, 1, 3, 2, 4, 5);  x_170 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_37: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_37.contiguous();  permute_37 = None
        windows_11: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_37.view(-1, 7, 7, 512);  contiguous_37 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_8: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_11.view(-1, 49, 512);  windows_11 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_34: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_8, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_bias_);  x_windows_8 = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_16: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_34.reshape(100, 49, 3, 16, 32);  linear_34 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_8: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_16.permute(2, 0, 3, 1, 4);  reshape_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_16: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_8[0]
        k_8: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_8[1]
        v_8: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_8[2];  qkv_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_17: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_16 * 0.1767766952966369;  q_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_19: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_8.transpose(-2, -1);  k_8 = None
        attn_40: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_17 @ transpose_19;  q_17 = transpose_19 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_108: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_51: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_parameters_relative_position_bias_table_[view_108];  l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_parameters_relative_position_bias_table_ = view_108 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_16: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_51.view(49, 49, -1);  getitem_51 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_39: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_16.permute(2, 0, 1);  relative_position_bias_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_17: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_39.contiguous();  permute_39 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_22: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_17.unsqueeze(0);  relative_position_bias_17 = None
        attn_41: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_40 + unsqueeze_22;  attn_40 = unsqueeze_22 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_42: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_41, -1, _stacklevel = 5);  attn_41 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_43: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_42, 0.0, False, False);  attn_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_17: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_43 @ v_8;  attn_43 = v_8 = None
        transpose_20: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_17.transpose(1, 2);  matmul_17 = None
        x_171: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_20.reshape(100, 49, 512);  transpose_20 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_172: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_171, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_bias_);  x_171 = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_173: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_172, 0.0, False, False);  x_172 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_8: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_173.view(-1, 7, 7, 512);  x_173 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_174: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_8.view(1, 10, 10, 7, 7, -1);  attn_windows_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_40: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_174.permute(0, 1, 3, 2, 4, 5);  x_174 = None
        contiguous_39: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_40.contiguous();  permute_40 = None
        x_175: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_39.view(1, 70, 70, -1);  contiguous_39 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_52: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_175[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_175 = None
        x_176: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_52.contiguous();  getitem_52 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_177: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_176.view(1, 4624, 512);  x_176 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_178: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_166 + x_177;  x_166 = x_177 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_22: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_178, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_179: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_22, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_22 = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_180: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_179, approximate = 'none');  x_179 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_181: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_180, 0.0, False, False);  x_180 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_182: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_181, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_bias_);  x_181 = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_4_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_183: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_182, 0.0, False, False);  x_182 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_184: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_178 + x_183;  x_178 = x_183 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_185: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_184, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_186: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_185.view(1, 68, 68, 512);  x_185 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_187: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_186, (0, 0, 0, 2, 0, 2), 'constant', None);  x_186 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_4: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_187, shifts = (-3, -3), dims = (1, 2));  x_187 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_188: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_4.view(1, 10, 7, 10, 7, 512);  shifted_x_4 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_41: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_188.permute(0, 1, 3, 2, 4, 5);  x_188 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_41: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_41.contiguous();  permute_41 = None
        windows_12: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_41.view(-1, 7, 7, 512);  contiguous_41 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_9: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_12.view(-1, 49, 512);  windows_12 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_38: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_9, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_bias_);  x_windows_9 = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_18: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_38.reshape(100, 49, 3, 16, 32);  linear_38 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_9: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_18.permute(2, 0, 3, 1, 4);  reshape_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_18: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_9[0]
        k_9: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_9[1]
        v_9: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_9[2];  qkv_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_19: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_18 * 0.1767766952966369;  q_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_21: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_9.transpose(-2, -1);  k_9 = None
        attn_44: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_19 @ transpose_21;  q_19 = transpose_21 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_118: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_56: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_parameters_relative_position_bias_table_[view_118];  l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_parameters_relative_position_bias_table_ = view_118 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_18: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_56.view(49, 49, -1);  getitem_56 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_43: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_18.permute(2, 0, 1);  relative_position_bias_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_19: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_43.contiguous();  permute_43 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_23: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_19.unsqueeze(0);  relative_position_bias_19 = None
        attn_45: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_44 + unsqueeze_23;  attn_44 = unsqueeze_23 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_120: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_45.view(1, 100, 16, 49, 49);  attn_45 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_24: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_25: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_24.unsqueeze(0);  unsqueeze_24 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_46: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_120 + unsqueeze_25;  view_120 = unsqueeze_25 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_47: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_46.view(-1, 16, 49, 49);  attn_46 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_48: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_47, -1, _stacklevel = 5);  attn_47 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_49: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_48, 0.0, False, False);  attn_48 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_19: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_49 @ v_9;  attn_49 = v_9 = None
        transpose_22: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_19.transpose(1, 2);  matmul_19 = None
        x_189: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_22.reshape(100, 49, 512);  transpose_22 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_190: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_189, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_bias_);  x_189 = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_191: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_190, 0.0, False, False);  x_190 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_9: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_191.view(-1, 7, 7, 512);  x_191 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_192: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_9.view(1, 10, 10, 7, 7, -1);  attn_windows_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_44: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_192.permute(0, 1, 3, 2, 4, 5);  x_192 = None
        contiguous_43: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_44.contiguous();  permute_44 = None
        x_193: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_43.view(1, 70, 70, -1);  contiguous_43 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_194: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_193, shifts = (3, 3), dims = (1, 2));  x_193 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_57: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_194[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_194 = None
        x_195: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_57.contiguous();  getitem_57 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_196: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_195.view(1, 4624, 512);  x_195 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_197: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_184 + x_196;  x_184 = x_196 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_24: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_197, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_198: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_24, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_24 = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_199: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_198, approximate = 'none');  x_198 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_200: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_199, 0.0, False, False);  x_199 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_201: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_200, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_bias_);  x_200 = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_5_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_202: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_201, 0.0, False, False);  x_201 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_203: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_197 + x_202;  x_197 = x_202 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_204: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_203, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_205: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_204.view(1, 68, 68, 512);  x_204 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_206: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_205, (0, 0, 0, 2, 0, 2), 'constant', None);  x_205 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_207: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_206.view(1, 10, 7, 10, 7, 512);  x_206 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_45: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_207.permute(0, 1, 3, 2, 4, 5);  x_207 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_45: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_45.contiguous();  permute_45 = None
        windows_13: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_45.view(-1, 7, 7, 512);  contiguous_45 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_10: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_13.view(-1, 49, 512);  windows_13 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_42: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_10, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_bias_);  x_windows_10 = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_20: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_42.reshape(100, 49, 3, 16, 32);  linear_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_10: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_20.permute(2, 0, 3, 1, 4);  reshape_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_20: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_10[0]
        k_10: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_10[1]
        v_10: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_10[2];  qkv_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_21: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_20 * 0.1767766952966369;  q_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_23: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_10.transpose(-2, -1);  k_10 = None
        attn_50: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_21 @ transpose_23;  q_21 = transpose_23 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_130: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_61: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_parameters_relative_position_bias_table_[view_130];  l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_parameters_relative_position_bias_table_ = view_130 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_20: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_61.view(49, 49, -1);  getitem_61 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_47: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_20.permute(2, 0, 1);  relative_position_bias_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_21: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_47.contiguous();  permute_47 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_26: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_21.unsqueeze(0);  relative_position_bias_21 = None
        attn_51: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_50 + unsqueeze_26;  attn_50 = unsqueeze_26 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_52: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_51, -1, _stacklevel = 5);  attn_51 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_53: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_52, 0.0, False, False);  attn_52 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_21: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_53 @ v_10;  attn_53 = v_10 = None
        transpose_24: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_21.transpose(1, 2);  matmul_21 = None
        x_208: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_24.reshape(100, 49, 512);  transpose_24 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_209: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_208, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_bias_);  x_208 = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_210: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_209, 0.0, False, False);  x_209 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_10: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_210.view(-1, 7, 7, 512);  x_210 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_211: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_10.view(1, 10, 10, 7, 7, -1);  attn_windows_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_48: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_211.permute(0, 1, 3, 2, 4, 5);  x_211 = None
        contiguous_47: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_48.contiguous();  permute_48 = None
        x_212: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_47.view(1, 70, 70, -1);  contiguous_47 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_62: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_212[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_212 = None
        x_213: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_62.contiguous();  getitem_62 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_214: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_213.view(1, 4624, 512);  x_213 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_215: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_203 + x_214;  x_203 = x_214 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_26: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_215, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_216: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_26, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_26 = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_217: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_216, approximate = 'none');  x_216 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_218: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_217, 0.0, False, False);  x_217 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_219: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_218, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_bias_);  x_218 = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_6_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_220: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_219, 0.0, False, False);  x_219 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_221: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_215 + x_220;  x_215 = x_220 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_222: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_221, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_223: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_222.view(1, 68, 68, 512);  x_222 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_224: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_223, (0, 0, 0, 2, 0, 2), 'constant', None);  x_223 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_5: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_224, shifts = (-3, -3), dims = (1, 2));  x_224 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_225: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_5.view(1, 10, 7, 10, 7, 512);  shifted_x_5 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_49: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_225.permute(0, 1, 3, 2, 4, 5);  x_225 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_49: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_49.contiguous();  permute_49 = None
        windows_14: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_49.view(-1, 7, 7, 512);  contiguous_49 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_11: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_14.view(-1, 49, 512);  windows_14 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_46: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_11, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_bias_);  x_windows_11 = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_22: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_46.reshape(100, 49, 3, 16, 32);  linear_46 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_11: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_22.permute(2, 0, 3, 1, 4);  reshape_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_22: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_11[0]
        k_11: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_11[1]
        v_11: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_11[2];  qkv_11 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_23: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_22 * 0.1767766952966369;  q_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_25: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_11.transpose(-2, -1);  k_11 = None
        attn_54: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_23 @ transpose_25;  q_23 = transpose_25 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_140: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_66: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_parameters_relative_position_bias_table_[view_140];  l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_parameters_relative_position_bias_table_ = view_140 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_22: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_66.view(49, 49, -1);  getitem_66 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_51: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_22.permute(2, 0, 1);  relative_position_bias_22 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_23: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_51.contiguous();  permute_51 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_27: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_23.unsqueeze(0);  relative_position_bias_23 = None
        attn_55: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_54 + unsqueeze_27;  attn_54 = unsqueeze_27 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_142: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_55.view(1, 100, 16, 49, 49);  attn_55 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_28: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_29: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_28.unsqueeze(0);  unsqueeze_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_56: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_142 + unsqueeze_29;  view_142 = unsqueeze_29 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_57: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_56.view(-1, 16, 49, 49);  attn_56 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_58: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_57, -1, _stacklevel = 5);  attn_57 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_59: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_58, 0.0, False, False);  attn_58 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_23: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_59 @ v_11;  attn_59 = v_11 = None
        transpose_26: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_23.transpose(1, 2);  matmul_23 = None
        x_226: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_26.reshape(100, 49, 512);  transpose_26 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_227: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_226, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_bias_);  x_226 = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_228: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_227, 0.0, False, False);  x_227 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_11: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_228.view(-1, 7, 7, 512);  x_228 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_229: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_11.view(1, 10, 10, 7, 7, -1);  attn_windows_11 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_52: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_229.permute(0, 1, 3, 2, 4, 5);  x_229 = None
        contiguous_51: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_52.contiguous();  permute_52 = None
        x_230: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_51.view(1, 70, 70, -1);  contiguous_51 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_231: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_230, shifts = (3, 3), dims = (1, 2));  x_230 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_67: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_231[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_231 = None
        x_232: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_67.contiguous();  getitem_67 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_233: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_232.view(1, 4624, 512);  x_232 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_234: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_221 + x_233;  x_221 = x_233 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_28: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_234, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_235: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_28, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_28 = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_236: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_235, approximate = 'none');  x_235 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_237: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_236, 0.0, False, False);  x_236 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_238: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_237, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_bias_);  x_237 = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_7_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_239: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_238, 0.0, False, False);  x_238 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_240: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_234 + x_239;  x_234 = x_239 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_241: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_240, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_242: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_241.view(1, 68, 68, 512);  x_241 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_243: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_242, (0, 0, 0, 2, 0, 2), 'constant', None);  x_242 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_244: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_243.view(1, 10, 7, 10, 7, 512);  x_243 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_53: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_244.permute(0, 1, 3, 2, 4, 5);  x_244 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_53: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_53.contiguous();  permute_53 = None
        windows_15: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_53.view(-1, 7, 7, 512);  contiguous_53 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_12: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_15.view(-1, 49, 512);  windows_15 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_50: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_12, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_bias_);  x_windows_12 = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_24: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_50.reshape(100, 49, 3, 16, 32);  linear_50 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_12: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_24.permute(2, 0, 3, 1, 4);  reshape_24 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_24: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_12[0]
        k_12: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_12[1]
        v_12: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_12[2];  qkv_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_25: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_24 * 0.1767766952966369;  q_24 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_27: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_12.transpose(-2, -1);  k_12 = None
        attn_60: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_25 @ transpose_27;  q_25 = transpose_27 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_152: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_71: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_parameters_relative_position_bias_table_[view_152];  l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_parameters_relative_position_bias_table_ = view_152 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_24: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_71.view(49, 49, -1);  getitem_71 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_55: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_24.permute(2, 0, 1);  relative_position_bias_24 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_25: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_55.contiguous();  permute_55 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_30: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_25.unsqueeze(0);  relative_position_bias_25 = None
        attn_61: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_60 + unsqueeze_30;  attn_60 = unsqueeze_30 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_62: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_61, -1, _stacklevel = 5);  attn_61 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_63: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_62, 0.0, False, False);  attn_62 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_25: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_63 @ v_12;  attn_63 = v_12 = None
        transpose_28: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_25.transpose(1, 2);  matmul_25 = None
        x_245: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_28.reshape(100, 49, 512);  transpose_28 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_246: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_245, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_bias_);  x_245 = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_247: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_246, 0.0, False, False);  x_246 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_12: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_247.view(-1, 7, 7, 512);  x_247 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_248: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_12.view(1, 10, 10, 7, 7, -1);  attn_windows_12 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_56: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_248.permute(0, 1, 3, 2, 4, 5);  x_248 = None
        contiguous_55: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_56.contiguous();  permute_56 = None
        x_249: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_55.view(1, 70, 70, -1);  contiguous_55 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_72: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_249[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_249 = None
        x_250: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_72.contiguous();  getitem_72 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_251: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_250.view(1, 4624, 512);  x_250 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_252: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_240 + x_251;  x_240 = x_251 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_30: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_252, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_253: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_30, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_30 = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_254: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_253, approximate = 'none');  x_253 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_255: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_254, 0.0, False, False);  x_254 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_256: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_255, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_bias_);  x_255 = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_8_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_257: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_256, 0.0, False, False);  x_256 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_258: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_252 + x_257;  x_252 = x_257 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_259: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_258, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_260: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_259.view(1, 68, 68, 512);  x_259 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_261: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_260, (0, 0, 0, 2, 0, 2), 'constant', None);  x_260 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_6: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_261, shifts = (-3, -3), dims = (1, 2));  x_261 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_262: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_6.view(1, 10, 7, 10, 7, 512);  shifted_x_6 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_57: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_262.permute(0, 1, 3, 2, 4, 5);  x_262 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_57: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_57.contiguous();  permute_57 = None
        windows_16: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_57.view(-1, 7, 7, 512);  contiguous_57 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_13: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_16.view(-1, 49, 512);  windows_16 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_54: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_13, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_bias_);  x_windows_13 = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_26: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_54.reshape(100, 49, 3, 16, 32);  linear_54 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_13: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_26.permute(2, 0, 3, 1, 4);  reshape_26 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_26: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_13[0]
        k_13: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_13[1]
        v_13: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_13[2];  qkv_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_27: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_26 * 0.1767766952966369;  q_26 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_29: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_13.transpose(-2, -1);  k_13 = None
        attn_64: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_27 @ transpose_29;  q_27 = transpose_29 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_162: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_76: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_parameters_relative_position_bias_table_[view_162];  l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_parameters_relative_position_bias_table_ = view_162 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_26: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_76.view(49, 49, -1);  getitem_76 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_59: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_26.permute(2, 0, 1);  relative_position_bias_26 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_27: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_59.contiguous();  permute_59 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_31: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_27.unsqueeze(0);  relative_position_bias_27 = None
        attn_65: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_64 + unsqueeze_31;  attn_64 = unsqueeze_31 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_164: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_65.view(1, 100, 16, 49, 49);  attn_65 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_32: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_33: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_32.unsqueeze(0);  unsqueeze_32 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_66: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_164 + unsqueeze_33;  view_164 = unsqueeze_33 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_67: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_66.view(-1, 16, 49, 49);  attn_66 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_68: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_67, -1, _stacklevel = 5);  attn_67 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_69: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_68, 0.0, False, False);  attn_68 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_27: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_69 @ v_13;  attn_69 = v_13 = None
        transpose_30: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_27.transpose(1, 2);  matmul_27 = None
        x_263: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_30.reshape(100, 49, 512);  transpose_30 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_264: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_263, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_bias_);  x_263 = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_265: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_264, 0.0, False, False);  x_264 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_13: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_265.view(-1, 7, 7, 512);  x_265 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_266: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_13.view(1, 10, 10, 7, 7, -1);  attn_windows_13 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_60: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_266.permute(0, 1, 3, 2, 4, 5);  x_266 = None
        contiguous_59: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_60.contiguous();  permute_60 = None
        x_267: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_59.view(1, 70, 70, -1);  contiguous_59 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_268: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_267, shifts = (3, 3), dims = (1, 2));  x_267 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_77: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_268[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_268 = None
        x_269: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_77.contiguous();  getitem_77 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_270: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_269.view(1, 4624, 512);  x_269 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_271: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_258 + x_270;  x_258 = x_270 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_32: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_271, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_272: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_32, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_32 = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_273: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_272, approximate = 'none');  x_272 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_274: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_273, 0.0, False, False);  x_273 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_275: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_274, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_bias_);  x_274 = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_9_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_276: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_275, 0.0, False, False);  x_275 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_277: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_271 + x_276;  x_271 = x_276 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_278: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_277, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_279: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_278.view(1, 68, 68, 512);  x_278 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_280: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_279, (0, 0, 0, 2, 0, 2), 'constant', None);  x_279 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_281: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_280.view(1, 10, 7, 10, 7, 512);  x_280 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_61: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_281.permute(0, 1, 3, 2, 4, 5);  x_281 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_61: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_61.contiguous();  permute_61 = None
        windows_17: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_61.view(-1, 7, 7, 512);  contiguous_61 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_14: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_17.view(-1, 49, 512);  windows_17 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_58: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_14, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_bias_);  x_windows_14 = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_28: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_58.reshape(100, 49, 3, 16, 32);  linear_58 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_14: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_28.permute(2, 0, 3, 1, 4);  reshape_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_28: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_14[0]
        k_14: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_14[1]
        v_14: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_14[2];  qkv_14 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_29: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_28 * 0.1767766952966369;  q_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_31: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_14.transpose(-2, -1);  k_14 = None
        attn_70: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_29 @ transpose_31;  q_29 = transpose_31 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_174: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_81: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_parameters_relative_position_bias_table_[view_174];  l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_parameters_relative_position_bias_table_ = view_174 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_28: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_81.view(49, 49, -1);  getitem_81 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_63: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_28.permute(2, 0, 1);  relative_position_bias_28 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_29: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_63.contiguous();  permute_63 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_34: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_29.unsqueeze(0);  relative_position_bias_29 = None
        attn_71: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_70 + unsqueeze_34;  attn_70 = unsqueeze_34 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_72: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_71, -1, _stacklevel = 5);  attn_71 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_73: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_72, 0.0, False, False);  attn_72 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_29: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_73 @ v_14;  attn_73 = v_14 = None
        transpose_32: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_29.transpose(1, 2);  matmul_29 = None
        x_282: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_32.reshape(100, 49, 512);  transpose_32 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_283: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_282, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_bias_);  x_282 = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_284: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_283, 0.0, False, False);  x_283 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_14: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_284.view(-1, 7, 7, 512);  x_284 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_285: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_14.view(1, 10, 10, 7, 7, -1);  attn_windows_14 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_64: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_285.permute(0, 1, 3, 2, 4, 5);  x_285 = None
        contiguous_63: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_64.contiguous();  permute_64 = None
        x_286: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_63.view(1, 70, 70, -1);  contiguous_63 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_82: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_286[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_286 = None
        x_287: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_82.contiguous();  getitem_82 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_288: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_287.view(1, 4624, 512);  x_287 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_289: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_277 + x_288;  x_277 = x_288 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_34: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_289, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_290: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_34, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_34 = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_291: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_290, approximate = 'none');  x_290 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_292: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_291, 0.0, False, False);  x_291 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_293: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_292, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_bias_);  x_292 = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_10_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_294: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_293, 0.0, False, False);  x_293 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_295: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_289 + x_294;  x_289 = x_294 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_296: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_295, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_297: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_296.view(1, 68, 68, 512);  x_296 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_298: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_297, (0, 0, 0, 2, 0, 2), 'constant', None);  x_297 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_7: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_298, shifts = (-3, -3), dims = (1, 2));  x_298 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_299: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_7.view(1, 10, 7, 10, 7, 512);  shifted_x_7 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_65: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_299.permute(0, 1, 3, 2, 4, 5);  x_299 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_65: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_65.contiguous();  permute_65 = None
        windows_18: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_65.view(-1, 7, 7, 512);  contiguous_65 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_15: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_18.view(-1, 49, 512);  windows_18 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_62: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_15, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_bias_);  x_windows_15 = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_30: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_62.reshape(100, 49, 3, 16, 32);  linear_62 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_15: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_30.permute(2, 0, 3, 1, 4);  reshape_30 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_30: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_15[0]
        k_15: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_15[1]
        v_15: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_15[2];  qkv_15 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_31: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_30 * 0.1767766952966369;  q_30 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_33: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_15.transpose(-2, -1);  k_15 = None
        attn_74: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_31 @ transpose_33;  q_31 = transpose_33 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_184: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_86: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_parameters_relative_position_bias_table_[view_184];  l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_parameters_relative_position_bias_table_ = view_184 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_30: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_86.view(49, 49, -1);  getitem_86 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_67: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_30.permute(2, 0, 1);  relative_position_bias_30 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_31: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_67.contiguous();  permute_67 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_35: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_31.unsqueeze(0);  relative_position_bias_31 = None
        attn_75: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_74 + unsqueeze_35;  attn_74 = unsqueeze_35 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_186: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_75.view(1, 100, 16, 49, 49);  attn_75 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_36: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_37: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_36.unsqueeze(0);  unsqueeze_36 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_76: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_186 + unsqueeze_37;  view_186 = unsqueeze_37 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_77: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_76.view(-1, 16, 49, 49);  attn_76 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_78: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_77, -1, _stacklevel = 5);  attn_77 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_79: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_78, 0.0, False, False);  attn_78 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_31: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_79 @ v_15;  attn_79 = v_15 = None
        transpose_34: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_31.transpose(1, 2);  matmul_31 = None
        x_300: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_34.reshape(100, 49, 512);  transpose_34 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_301: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_300, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_bias_);  x_300 = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_302: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_301, 0.0, False, False);  x_301 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_15: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_302.view(-1, 7, 7, 512);  x_302 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_303: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_15.view(1, 10, 10, 7, 7, -1);  attn_windows_15 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_68: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_303.permute(0, 1, 3, 2, 4, 5);  x_303 = None
        contiguous_67: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_68.contiguous();  permute_68 = None
        x_304: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_67.view(1, 70, 70, -1);  contiguous_67 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_305: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_304, shifts = (3, 3), dims = (1, 2));  x_304 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_87: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_305[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_305 = None
        x_306: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_87.contiguous();  getitem_87 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_307: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_306.view(1, 4624, 512);  x_306 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_308: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_295 + x_307;  x_295 = x_307 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_36: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_308, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_309: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_36, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_36 = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_310: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_309, approximate = 'none');  x_309 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_311: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_310, 0.0, False, False);  x_310 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_312: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_311, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_bias_);  x_311 = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_11_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_313: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_312, 0.0, False, False);  x_312 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_314: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_308 + x_313;  x_308 = x_313 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_315: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_314, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_316: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_315.view(1, 68, 68, 512);  x_315 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_317: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_316, (0, 0, 0, 2, 0, 2), 'constant', None);  x_316 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_318: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_317.view(1, 10, 7, 10, 7, 512);  x_317 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_69: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_318.permute(0, 1, 3, 2, 4, 5);  x_318 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_69: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_69.contiguous();  permute_69 = None
        windows_19: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_69.view(-1, 7, 7, 512);  contiguous_69 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_16: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_19.view(-1, 49, 512);  windows_19 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_66: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_16, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_bias_);  x_windows_16 = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_32: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_66.reshape(100, 49, 3, 16, 32);  linear_66 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_16: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_32.permute(2, 0, 3, 1, 4);  reshape_32 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_32: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_16[0]
        k_16: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_16[1]
        v_16: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_16[2];  qkv_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_33: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_32 * 0.1767766952966369;  q_32 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_35: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_16.transpose(-2, -1);  k_16 = None
        attn_80: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_33 @ transpose_35;  q_33 = transpose_35 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_196: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_91: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_parameters_relative_position_bias_table_[view_196];  l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_parameters_relative_position_bias_table_ = view_196 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_32: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_91.view(49, 49, -1);  getitem_91 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_71: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_32.permute(2, 0, 1);  relative_position_bias_32 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_33: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_71.contiguous();  permute_71 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_38: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_33.unsqueeze(0);  relative_position_bias_33 = None
        attn_81: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_80 + unsqueeze_38;  attn_80 = unsqueeze_38 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_82: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_81, -1, _stacklevel = 5);  attn_81 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_83: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_82, 0.0, False, False);  attn_82 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_33: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_83 @ v_16;  attn_83 = v_16 = None
        transpose_36: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_33.transpose(1, 2);  matmul_33 = None
        x_319: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_36.reshape(100, 49, 512);  transpose_36 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_320: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_319, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_bias_);  x_319 = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_321: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_320, 0.0, False, False);  x_320 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_16: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_321.view(-1, 7, 7, 512);  x_321 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_322: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_16.view(1, 10, 10, 7, 7, -1);  attn_windows_16 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_72: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_322.permute(0, 1, 3, 2, 4, 5);  x_322 = None
        contiguous_71: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_72.contiguous();  permute_72 = None
        x_323: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_71.view(1, 70, 70, -1);  contiguous_71 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_92: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_323[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_323 = None
        x_324: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_92.contiguous();  getitem_92 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_325: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_324.view(1, 4624, 512);  x_324 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_326: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_314 + x_325;  x_314 = x_325 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_38: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_326, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_327: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_38, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_38 = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_328: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_327, approximate = 'none');  x_327 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_329: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_328, 0.0, False, False);  x_328 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_330: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_329, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_bias_);  x_329 = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_12_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_331: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_330, 0.0, False, False);  x_330 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_332: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_326 + x_331;  x_326 = x_331 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_333: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_332, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_334: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_333.view(1, 68, 68, 512);  x_333 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_335: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_334, (0, 0, 0, 2, 0, 2), 'constant', None);  x_334 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_8: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_335, shifts = (-3, -3), dims = (1, 2));  x_335 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_336: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_8.view(1, 10, 7, 10, 7, 512);  shifted_x_8 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_73: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_336.permute(0, 1, 3, 2, 4, 5);  x_336 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_73: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_73.contiguous();  permute_73 = None
        windows_20: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_73.view(-1, 7, 7, 512);  contiguous_73 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_17: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_20.view(-1, 49, 512);  windows_20 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_70: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_17, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_bias_);  x_windows_17 = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_34: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_70.reshape(100, 49, 3, 16, 32);  linear_70 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_17: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_34.permute(2, 0, 3, 1, 4);  reshape_34 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_34: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_17[0]
        k_17: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_17[1]
        v_17: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_17[2];  qkv_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_35: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_34 * 0.1767766952966369;  q_34 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_37: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_17.transpose(-2, -1);  k_17 = None
        attn_84: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_35 @ transpose_37;  q_35 = transpose_37 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_206: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_96: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_parameters_relative_position_bias_table_[view_206];  l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_parameters_relative_position_bias_table_ = view_206 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_34: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_96.view(49, 49, -1);  getitem_96 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_75: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_34.permute(2, 0, 1);  relative_position_bias_34 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_35: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_75.contiguous();  permute_75 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_39: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_35.unsqueeze(0);  relative_position_bias_35 = None
        attn_85: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_84 + unsqueeze_39;  attn_84 = unsqueeze_39 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_208: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_85.view(1, 100, 16, 49, 49);  attn_85 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_40: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_41: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_40.unsqueeze(0);  unsqueeze_40 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_86: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_208 + unsqueeze_41;  view_208 = unsqueeze_41 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_87: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_86.view(-1, 16, 49, 49);  attn_86 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_88: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_87, -1, _stacklevel = 5);  attn_87 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_89: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_88, 0.0, False, False);  attn_88 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_35: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_89 @ v_17;  attn_89 = v_17 = None
        transpose_38: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_35.transpose(1, 2);  matmul_35 = None
        x_337: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_38.reshape(100, 49, 512);  transpose_38 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_338: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_337, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_bias_);  x_337 = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_339: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_338, 0.0, False, False);  x_338 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_17: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_339.view(-1, 7, 7, 512);  x_339 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_340: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_17.view(1, 10, 10, 7, 7, -1);  attn_windows_17 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_76: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_340.permute(0, 1, 3, 2, 4, 5);  x_340 = None
        contiguous_75: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_76.contiguous();  permute_76 = None
        x_341: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_75.view(1, 70, 70, -1);  contiguous_75 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_342: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_341, shifts = (3, 3), dims = (1, 2));  x_341 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_97: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_342[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_342 = None
        x_343: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_97.contiguous();  getitem_97 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_344: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_343.view(1, 4624, 512);  x_343 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_345: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_332 + x_344;  x_332 = x_344 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_40: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_345, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_346: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_40, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_40 = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_347: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_346, approximate = 'none');  x_346 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_348: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_347, 0.0, False, False);  x_347 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_349: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_348, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_bias_);  x_348 = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_13_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_350: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_349, 0.0, False, False);  x_349 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_351: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_345 + x_350;  x_345 = x_350 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_352: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_351, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_353: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_352.view(1, 68, 68, 512);  x_352 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_354: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_353, (0, 0, 0, 2, 0, 2), 'constant', None);  x_353 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_355: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_354.view(1, 10, 7, 10, 7, 512);  x_354 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_77: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_355.permute(0, 1, 3, 2, 4, 5);  x_355 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_77: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_77.contiguous();  permute_77 = None
        windows_21: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_77.view(-1, 7, 7, 512);  contiguous_77 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_18: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_21.view(-1, 49, 512);  windows_21 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_74: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_18, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_bias_);  x_windows_18 = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_36: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_74.reshape(100, 49, 3, 16, 32);  linear_74 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_18: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_36.permute(2, 0, 3, 1, 4);  reshape_36 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_36: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_18[0]
        k_18: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_18[1]
        v_18: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_18[2];  qkv_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_37: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_36 * 0.1767766952966369;  q_36 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_39: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_18.transpose(-2, -1);  k_18 = None
        attn_90: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_37 @ transpose_39;  q_37 = transpose_39 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_218: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_101: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_parameters_relative_position_bias_table_[view_218];  l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_parameters_relative_position_bias_table_ = view_218 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_36: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_101.view(49, 49, -1);  getitem_101 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_79: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_36.permute(2, 0, 1);  relative_position_bias_36 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_37: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_79.contiguous();  permute_79 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_42: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_37.unsqueeze(0);  relative_position_bias_37 = None
        attn_91: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_90 + unsqueeze_42;  attn_90 = unsqueeze_42 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_92: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_91, -1, _stacklevel = 5);  attn_91 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_93: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_92, 0.0, False, False);  attn_92 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_37: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_93 @ v_18;  attn_93 = v_18 = None
        transpose_40: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_37.transpose(1, 2);  matmul_37 = None
        x_356: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_40.reshape(100, 49, 512);  transpose_40 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_357: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_356, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_bias_);  x_356 = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_358: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_357, 0.0, False, False);  x_357 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_18: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_358.view(-1, 7, 7, 512);  x_358 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_359: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_18.view(1, 10, 10, 7, 7, -1);  attn_windows_18 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_80: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_359.permute(0, 1, 3, 2, 4, 5);  x_359 = None
        contiguous_79: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_80.contiguous();  permute_80 = None
        x_360: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_79.view(1, 70, 70, -1);  contiguous_79 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_102: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_360[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_360 = None
        x_361: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_102.contiguous();  getitem_102 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_362: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_361.view(1, 4624, 512);  x_361 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_363: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_351 + x_362;  x_351 = x_362 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_42: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_363, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_364: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_42, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_42 = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_365: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_364, approximate = 'none');  x_364 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_366: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_365, 0.0, False, False);  x_365 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_367: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_366, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_bias_);  x_366 = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_14_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_368: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_367, 0.0, False, False);  x_367 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_369: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_363 + x_368;  x_363 = x_368 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_370: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_369, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_371: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_370.view(1, 68, 68, 512);  x_370 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_372: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_371, (0, 0, 0, 2, 0, 2), 'constant', None);  x_371 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_9: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_372, shifts = (-3, -3), dims = (1, 2));  x_372 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_373: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_9.view(1, 10, 7, 10, 7, 512);  shifted_x_9 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_81: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_373.permute(0, 1, 3, 2, 4, 5);  x_373 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_81: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_81.contiguous();  permute_81 = None
        windows_22: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_81.view(-1, 7, 7, 512);  contiguous_81 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_19: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_22.view(-1, 49, 512);  windows_22 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_78: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_19, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_bias_);  x_windows_19 = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_38: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_78.reshape(100, 49, 3, 16, 32);  linear_78 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_19: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_38.permute(2, 0, 3, 1, 4);  reshape_38 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_38: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_19[0]
        k_19: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_19[1]
        v_19: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_19[2];  qkv_19 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_39: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_38 * 0.1767766952966369;  q_38 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_41: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_19.transpose(-2, -1);  k_19 = None
        attn_94: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_39 @ transpose_41;  q_39 = transpose_41 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_228: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_106: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_parameters_relative_position_bias_table_[view_228];  l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_parameters_relative_position_bias_table_ = view_228 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_38: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_106.view(49, 49, -1);  getitem_106 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_83: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_38.permute(2, 0, 1);  relative_position_bias_38 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_39: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_83.contiguous();  permute_83 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_43: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_39.unsqueeze(0);  relative_position_bias_39 = None
        attn_95: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_94 + unsqueeze_43;  attn_94 = unsqueeze_43 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_230: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_95.view(1, 100, 16, 49, 49);  attn_95 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_44: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1)
        unsqueeze_45: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_44.unsqueeze(0);  unsqueeze_44 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_96: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_230 + unsqueeze_45;  view_230 = unsqueeze_45 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_97: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_96.view(-1, 16, 49, 49);  attn_96 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_98: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_97, -1, _stacklevel = 5);  attn_97 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_99: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_98, 0.0, False, False);  attn_98 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_39: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_99 @ v_19;  attn_99 = v_19 = None
        transpose_42: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_39.transpose(1, 2);  matmul_39 = None
        x_374: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_42.reshape(100, 49, 512);  transpose_42 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_375: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_374, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_bias_);  x_374 = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_376: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_375, 0.0, False, False);  x_375 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_19: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_376.view(-1, 7, 7, 512);  x_376 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_377: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_19.view(1, 10, 10, 7, 7, -1);  attn_windows_19 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_84: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_377.permute(0, 1, 3, 2, 4, 5);  x_377 = None
        contiguous_83: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_84.contiguous();  permute_84 = None
        x_378: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_83.view(1, 70, 70, -1);  contiguous_83 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_379: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_378, shifts = (3, 3), dims = (1, 2));  x_378 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_107: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_379[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_379 = None
        x_380: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_107.contiguous();  getitem_107 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_381: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_380.view(1, 4624, 512);  x_380 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_382: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_369 + x_381;  x_369 = x_381 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_44: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_382, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_383: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_44, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_44 = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_384: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_383, approximate = 'none');  x_383 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_385: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_384, 0.0, False, False);  x_384 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_386: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_385, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_bias_);  x_385 = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_15_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_387: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_386, 0.0, False, False);  x_386 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_388: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_382 + x_387;  x_382 = x_387 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_389: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_388, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_390: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_389.view(1, 68, 68, 512);  x_389 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_391: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_390, (0, 0, 0, 2, 0, 2), 'constant', None);  x_390 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_392: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = x_391.view(1, 10, 7, 10, 7, 512);  x_391 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_85: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_392.permute(0, 1, 3, 2, 4, 5);  x_392 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_85: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_85.contiguous();  permute_85 = None
        windows_23: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_85.view(-1, 7, 7, 512);  contiguous_85 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_20: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_23.view(-1, 49, 512);  windows_23 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_82: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_20, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_bias_);  x_windows_20 = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_40: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_82.reshape(100, 49, 3, 16, 32);  linear_82 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_20: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_40.permute(2, 0, 3, 1, 4);  reshape_40 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_40: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_20[0]
        k_20: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_20[1]
        v_20: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_20[2];  qkv_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_41: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_40 * 0.1767766952966369;  q_40 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_43: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_20.transpose(-2, -1);  k_20 = None
        attn_100: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_41 @ transpose_43;  q_41 = transpose_43 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_240: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_111: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_parameters_relative_position_bias_table_[view_240];  l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_parameters_relative_position_bias_table_ = view_240 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_40: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_111.view(49, 49, -1);  getitem_111 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_87: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_40.permute(2, 0, 1);  relative_position_bias_40 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_41: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_87.contiguous();  permute_87 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_46: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_41.unsqueeze(0);  relative_position_bias_41 = None
        attn_101: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_100 + unsqueeze_46;  attn_100 = unsqueeze_46 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_102: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_101, -1, _stacklevel = 5);  attn_101 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_103: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_102, 0.0, False, False);  attn_102 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_41: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_103 @ v_20;  attn_103 = v_20 = None
        transpose_44: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_41.transpose(1, 2);  matmul_41 = None
        x_393: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_44.reshape(100, 49, 512);  transpose_44 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_394: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_393, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_bias_);  x_393 = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_395: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_394, 0.0, False, False);  x_394 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_20: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_395.view(-1, 7, 7, 512);  x_395 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_396: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_20.view(1, 10, 10, 7, 7, -1);  attn_windows_20 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_88: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_396.permute(0, 1, 3, 2, 4, 5);  x_396 = None
        contiguous_87: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_88.contiguous();  permute_88 = None
        x_397: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_87.view(1, 70, 70, -1);  contiguous_87 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_112: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_397[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_397 = None
        x_398: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_112.contiguous();  getitem_112 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_399: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_398.view(1, 4624, 512);  x_398 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_400: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_388 + x_399;  x_388 = x_399 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_46: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_400, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_401: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_46, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_46 = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_402: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_401, approximate = 'none');  x_401 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_403: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_402, 0.0, False, False);  x_402 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_404: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_403, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_bias_);  x_403 = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_16_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_405: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_404, 0.0, False, False);  x_404 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_406: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_400 + x_405;  x_400 = x_405 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_407: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_406, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm1_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:269 in forward, code: x = x.view(B, H, W, C)
        x_408: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_407.view(1, 68, 68, 512);  x_407 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
        x_409: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch._C._nn.pad(x_408, (0, 0, 0, 2, 0, 2), 'constant', None);  x_408 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:280 in forward, code: shifted_x = torch.roll(x,
        shifted_x_10: "f32[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_409, shifts = (-3, -3), dims = (1, 2));  x_409 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:75 in window_partition, code: x = x.view(B, H // window_size, window_size, W // window_size, window_size,
        x_410: "f32[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = shifted_x_10.view(1, 10, 7, 10, 7, 512);  shifted_x_10 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:77 in window_partition, code: windows = x.permute(0, 1, 3, 2, 4,
        permute_89: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 3584, 35840, 512, 1]cuda:0" = x_410.permute(0, 1, 3, 2, 4, 5);  x_410 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:78 in window_partition, code: 5).contiguous().view(-1, window_size, window_size, C)
        contiguous_89: "f32[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = permute_89.contiguous();  permute_89 = None
        windows_24: "f32[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = contiguous_89.view(-1, 7, 7, 512);  contiguous_89 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:291 in forward, code: x_windows = x_windows.view(-1, self.window_size * self.window_size,
        x_windows_21: "f32[100, 49, 512][25088, 512, 1]cuda:0" = windows_24.view(-1, 49, 512);  windows_24 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        linear_86: "f16[100, 49, 1536][75264, 1536, 1]cuda:0" = torch._C._nn.linear(x_windows_21, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_bias_);  x_windows_21 = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_qkv_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:164 in forward, code: qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads,
        reshape_42: "f16[100, 49, 3, 16, 32][75264, 1536, 512, 32, 1]cuda:0" = linear_86.reshape(100, 49, 3, 16, 32);  linear_86 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:165 in forward, code: C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv_21: "f16[3, 100, 16, 49, 32][512, 75264, 32, 1536, 1]cuda:0" = reshape_42.permute(2, 0, 3, 1, 4);  reshape_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:166 in forward, code: q, k, v = qkv[0], qkv[1], qkv[
        q_42: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_21[0]
        k_21: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_21[1]
        v_21: "f16[100, 16, 49, 32][75264, 32, 1536, 1]cuda:0" = qkv_21[2];  qkv_21 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:169 in forward, code: q = q * self.scale
        q_43: "f16[100, 16, 49, 32][25088, 32, 512, 1]cuda:0" = q_42 * 0.1767766952966369;  q_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:170 in forward, code: attn = (q @ k.transpose(-2, -1))
        transpose_45: "f16[100, 16, 32, 49][75264, 32, 1, 1536]cuda:0" = k_21.transpose(-2, -1);  k_21 = None
        attn_104: "f16[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = q_43 @ transpose_45;  q_43 = transpose_45 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        view_250: "i64[2401][1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_buffers_relative_position_index_.view(-1);  l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_buffers_relative_position_index_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:172 in forward, code: relative_position_bias = self.relative_position_bias_table[
        getitem_116: "f32[2401, 16][16, 1]cuda:0" = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_parameters_relative_position_bias_table_[view_250];  l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_parameters_relative_position_bias_table_ = view_250 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:173 in forward, code: self.relative_position_index.view(-1)].view(
        relative_position_bias_42: "f32[49, 49, 16][784, 16, 1]cuda:0" = getitem_116.view(49, 49, -1);  getitem_116 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:177 in forward, code: relative_position_bias = relative_position_bias.permute(
        permute_91: "f32[16, 49, 49][1, 784, 16]cuda:0" = relative_position_bias_42.permute(2, 0, 1);  relative_position_bias_42 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:178 in forward, code: 2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        relative_position_bias_43: "f32[16, 49, 49][2401, 49, 1]cuda:0" = permute_91.contiguous();  permute_91 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:179 in forward, code: attn = attn + relative_position_bias.unsqueeze(0)
        unsqueeze_47: "f32[1, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = relative_position_bias_43.unsqueeze(0);  relative_position_bias_43 = None
        attn_105: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_104 + unsqueeze_47;  attn_104 = unsqueeze_47 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        view_252: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = attn_105.view(1, 100, 16, 49, 49);  attn_105 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:184 in forward, code: N) + mask.unsqueeze(1).unsqueeze(0)
        unsqueeze_48: "f32[100, 1, 49, 49][2401, 2401, 49, 1]cuda:0" = attn_mask_5.unsqueeze(1);  attn_mask_5 = None
        unsqueeze_49: "f32[1, 100, 1, 49, 49][240100, 2401, 2401, 49, 1]cuda:0" = unsqueeze_48.unsqueeze(0);  unsqueeze_48 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:183 in forward, code: attn = attn.view(B_ // nW, nW, self.num_heads, N,
        attn_106: "f32[1, 100, 16, 49, 49][3841600, 38416, 2401, 49, 1]cuda:0" = view_252 + unsqueeze_49;  view_252 = unsqueeze_49 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:185 in forward, code: attn = attn.view(-1, self.num_heads, N, N)
        attn_107: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = attn_106.view(-1, 16, 49, 49);  attn_106 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:1553 in forward, code: return F.softmax(input, self.dim, _stacklevel=5)
        attn_108: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.softmax(attn_107, -1, _stacklevel = 5);  attn_107 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        attn_109: "f32[100, 16, 49, 49][38416, 2401, 49, 1]cuda:0" = torch.nn.functional.dropout(attn_108, 0.0, False, False);  attn_108 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:192 in forward, code: x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        matmul_43: "f16[100, 16, 49, 32][25088, 1568, 32, 1]cuda:0" = attn_109 @ v_21;  attn_109 = v_21 = None
        transpose_46: "f16[100, 49, 16, 32][25088, 32, 1568, 1]cuda:0" = matmul_43.transpose(1, 2);  matmul_43 = None
        x_411: "f16[100, 49, 512][25088, 512, 1]cuda:0" = transpose_46.reshape(100, 49, 512);  transpose_46 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_412: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch._C._nn.linear(x_411, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_bias_);  x_411 = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_attn_modules_proj_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_413: "f16[100, 49, 512][25088, 512, 1]cuda:0" = torch.nn.functional.dropout(x_412, 0.0, False, False);  x_412 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:299 in forward, code: attn_windows = attn_windows.view(-1, self.window_size,
        attn_windows_21: "f16[100, 7, 7, 512][25088, 3584, 512, 1]cuda:0" = x_413.view(-1, 7, 7, 512);  x_413 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:93 in window_reverse, code: x = windows.view(B, H // window_size, W // window_size, window_size,
        x_414: "f16[1, 10, 10, 7, 7, 512][2508800, 250880, 25088, 3584, 512, 1]cuda:0" = attn_windows_21.view(1, 10, 10, 7, 7, -1);  attn_windows_21 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:95 in window_reverse, code: x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
        permute_92: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 3584, 25088, 512, 1]cuda:0" = x_414.permute(0, 1, 3, 2, 4, 5);  x_414 = None
        contiguous_91: "f16[1, 10, 7, 10, 7, 512][2508800, 250880, 35840, 3584, 512, 1]cuda:0" = permute_92.contiguous();  permute_92 = None
        x_415: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = contiguous_91.view(1, 70, 70, -1);  contiguous_91 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:306 in forward, code: x = torch.roll(shifted_x,
        x_416: "f16[1, 70, 70, 512][2508800, 35840, 512, 1]cuda:0" = torch.roll(x_415, shifts = (3, 3), dims = (1, 2));  x_415 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:313 in forward, code: x = x[:, :H, :W, :].contiguous()
        getitem_117: "f16[1, 68, 68, 512][2508800, 35840, 512, 1]cuda:0" = x_416[(slice(None, None, None), slice(None, 68, None), slice(None, 68, None), slice(None, None, None))];  x_416 = None
        x_417: "f16[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = getitem_117.contiguous();  getitem_117 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:315 in forward, code: x = x.view(B, H * W, C)
        x_418: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_417.view(1, 4624, 512);  x_417 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:318 in forward, code: x = shortcut + self.drop_path(x)
        x_419: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_406 + x_418;  x_406 = x_418 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        layer_norm_48: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_419, (512,), l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_bias_, 1e-05);  l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_norm2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_420: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.linear(layer_norm_48, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_bias_);  layer_norm_48 = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc1_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/activation.py:704 in forward, code: return F.gelu(input, approximate=self.approximate)
        x_421: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch._C._nn.gelu(x_420, approximate = 'none');  x_420 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_422: "f16[1, 4624, 2048][9469952, 2048, 1]cuda:0" = torch.nn.functional.dropout(x_421, 0.0, False, False);  x_421 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/linear.py:116 in forward, code: return F.linear(input, self.weight, self.bias)
        x_423: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch._C._nn.linear(x_422, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_weight_, l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_bias_);  x_422 = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_weight_ = l_self_modules_layers_modules_2_modules_blocks_modules_17_modules_mlp_modules_fc2_parameters_bias_ = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/dropout.py:59 in forward, code: return F.dropout(input, self.p, self.training, self.inplace)
        x_424: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.dropout(x_423, 0.0, False, False);  x_423 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:319 in forward, code: x = x + self.drop_path(self.mlp(self.norm2(x)))
        x_425: "f16[1, 4624, 512][2367488, 512, 1]cuda:0" = x_419 + x_424;  x_419 = x_424 = None
        
        # File: /opt/conda/lib/python3.11/site-packages/torch/nn/modules/normalization.py:201 in forward, code: return F.layer_norm(
        x_out_2: "f32[1, 4624, 512][2367488, 512, 1]cuda:0" = torch.nn.functional.layer_norm(x_425, (512,), l_self_modules_norm2_parameters_weight_, l_self_modules_norm2_parameters_bias_, 1e-05);  x_425 = l_self_modules_norm2_parameters_weight_ = l_self_modules_norm2_parameters_bias_ = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:715 in forward, code: out = x_out.view(-1, H, W,
        view_258: "f32[1, 68, 68, 512][2367488, 34816, 512, 1]cuda:0" = x_out_2.view(-1, 68, 68, 512);  x_out_2 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:716 in forward, code: self.num_features[i]).permute(0, 3, 1,
        permute_93: "f32[1, 512, 68, 68][2367488, 1, 34816, 512]cuda:0" = view_258.permute(0, 3, 1, 2);  view_258 = None
        
        # File: /workspace/networks/encoders/swin/swin_transformer.py:717 in forward, code: 2).contiguous()
        out_2: "f32[1, 512, 68, 68][2367488, 4624, 68, 1]cuda:0" = permute_93.contiguous();  permute_93 = None
        return (out, out_1, out_2)