class <lambda>(torch.nn.Module):
    def forward(self, arg0_1: "Sym(s0)", arg1_1: "Sym(s1)", arg2_1: "f16[1, 15, 15, s0, s1]", arg3_1: "Sym(s10)", arg4_1: "Sym(s11)", arg5_1: "f16[1, 1, 225, s0*s1]", arg6_1: "f32[1, 1, 225, s0*s1]", arg7_1: "f32[1, 256, 225]", arg8_1: "Sym(s6)", arg9_1: "Sym(s7)", arg10_1: "Sym(s8)", arg11_1: "Sym(s9)", arg12_1: "b8[1, 1, s6, s7, s8]", arg13_1: "Sym(s10)", arg14_1: "Sym(s11)", arg15_1: "f16[1, 1, 256, s10*s11]", arg16_1: "f32[256, 256]", arg17_1: "f32[256]"):
        # 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)
        convert_element_type_9: "f16[256]" = torch.ops.prims.convert_element_type.default(arg17_1, torch.float16);  arg17_1 = None
        
        # File: /workspace/networks/layers/attention.py:346 in torch_dynamo_resume_in_forward_at_346, code: qk = self.correlation_sampler(q, k).view(
        mul: "Sym(s10*s11)" = arg3_1 * arg4_1
        
        # File: /workspace/networks/layers/attention.py:416 in local2global, code: global_attn = torch.zeros(
        add_1: "Sym(s10 + 14)" = arg3_1 + 14;  arg3_1 = None
        add_2: "Sym(s11 + 14)" = arg4_1 + 14;  arg4_1 = None
        full: "f32[1, 1, s10*s11, s10 + 14, s11 + 14]" = torch.ops.aten.full.default([1, 1, mul, add_1, add_2], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False);  add_1 = add_2 = None
        
        # File: /workspace/networks/layers/attention.py:419 in local2global, code: global_attn[local_mask.expand(batch_size, self.num_head,
        expand: "b8[1, 1, s6, s7, s8]" = torch.ops.aten.expand.default(arg12_1, [1, 1, -1, -1, -1]);  arg12_1 = None
        
        # File: /workspace/networks/layers/attention.py:346 in torch_dynamo_resume_in_forward_at_346, code: qk = self.correlation_sampler(q, k).view(
        view: "f16[1, 1, 225, s0*s1]" = torch.ops.aten.reshape.default(arg2_1, [1, 1, 225, mul]);  arg2_1 = None
        
        # 
        add: "f16[1, 1, 225, s0*s1]" = torch.ops.aten.add.Tensor(view, arg5_1);  arg5_1 = None
        
        # 
        mul_1: "f32[1, 1, 225, s0*s1]" = torch.ops.aten.mul.Tensor(arg6_1, 10000.0);  arg6_1 = None
        sub: "f32[1, 1, 225, s0*s1]" = torch.ops.aten.sub.Tensor(add, mul_1);  add = mul_1 = None
        
        # 
        amax: "f32[1, 1, 1, s0*s1]" = torch.ops.aten.amax.default(sub, [2], True)
        sub_1: "f32[1, 1, 225, s0*s1]" = torch.ops.aten.sub.Tensor(sub, amax);  sub = amax = None
        exp: "f32[1, 1, 225, s0*s1]" = torch.ops.aten.exp.default(sub_1);  sub_1 = None
        sum_1: "f32[1, 1, 1, s0*s1]" = torch.ops.aten.sum.dim_IntList(exp, [2], True)
        div: "f32[1, 1, 225, s0*s1]" = torch.ops.aten.div.Tensor(exp, sum_1);  exp = sum_1 = None
        
        # File: /workspace/networks/layers/attention.py:420 in local2global, code: -1, -1, -1)] = local_attn.transpose(
        permute_5: "f32[1, 1, s0*s1, 225]" = torch.ops.aten.permute.default(div, [0, 1, 3, 2])
        
        # File: /workspace/networks/layers/attention.py:421 in local2global, code: -1, -2).reshape(-1)
        clone_1: "f32[1, 1, s0*s1, 225]" = torch.ops.aten.clone.default(permute_5, memory_format = torch.contiguous_format);  permute_5 = None
        
        # File: /workspace/networks/layers/attention.py:370 in torch_dynamo_resume_in_forward_at_346, code: self.relative_emb_v)
        sym_size_int: "Sym(s0*s1)" = torch.ops.aten.sym_size.int(view, 3);  view = None
        
        # File: /workspace/networks/layers/attention.py:421 in local2global, code: -1, -2).reshape(-1)
        mul_3: "Sym(225*s0*s1)" = sym_size_int * 225
        floordiv: "Sym(225*s0*s1)" = mul_3 // 1;  mul_3 = None
        view_5: "f32[225*s0*s1]" = torch.ops.aten.reshape.default(clone_1, [floordiv]);  clone_1 = floordiv = None
        
        # File: /workspace/networks/layers/attention.py:419 in local2global, code: global_attn[local_mask.expand(batch_size, self.num_head,
        index_put: "f32[1, 1, s10*s11, s10 + 14, s11 + 14]" = torch.ops.aten.index_put_.default(full, [expand], view_5);  full = expand = view_5 = None
        
        # File: /workspace/networks/layers/attention.py:423 in local2global, code: self.max_dis:-self.max_dis].reshape(
        slice_9: "f32[1, 1, s10*s11, s10, s11 + 14]" = torch.ops.aten.slice.Tensor(index_put, 3, 7, -7);  index_put = None
        slice_10: "f32[1, 1, s10*s11, s10, s11]" = torch.ops.aten.slice.Tensor(slice_9, 4, 7, -7);  slice_9 = None
        clone_2: "f32[1, 1, s10*s11, s10, s11]" = torch.ops.aten.clone.default(slice_10, memory_format = torch.contiguous_format);  slice_10 = None
        view_6: "f32[1, 1, s10*s11, s10*s11]" = torch.ops.aten.reshape.default(clone_2, [1, 1, mul, mul]);  clone_2 = None
        
        # 
        convert_element_type_6: "f16[1, 1, s10*s11, s10*s11]" = torch.ops.prims.convert_element_type.default(view_6, torch.float16);  view_6 = None
        expand_1: "f16[1, 1, s10*s11, s10*s11]" = torch.ops.aten.expand.default(convert_element_type_6, [1, 1, mul, mul]);  convert_element_type_6 = None
        view_7: "f16[1, s10*s11, s10*s11]" = torch.ops.aten.reshape.default(expand_1, [1, mul, mul]);  expand_1 = None
        permute_6: "f16[1, 1, s10*s11, 256]" = torch.ops.aten.permute.default(arg15_1, [0, 1, 3, 2]);  arg15_1 = None
        expand_2: "f16[1, 1, s10*s11, 256]" = torch.ops.aten.expand.default(permute_6, [1, 1, mul, 256]);  permute_6 = None
        view_8: "f16[1, s10*s11, 256]" = torch.ops.aten.reshape.default(expand_2, [1, mul, 256]);  expand_2 = None
        bmm_1: "f16[1, s10*s11, 256]" = torch.ops.aten.bmm.default(view_7, view_8);  view_7 = view_8 = None
        view_9: "f16[1, 1, s10*s11, 256]" = torch.ops.aten.reshape.default(bmm_1, [1, 1, mul, 256]);  bmm_1 = None
        
        # File: /workspace/networks/layers/attention.py:370 in torch_dynamo_resume_in_forward_at_346, code: self.relative_emb_v)
        convert_element_type_2: "f16[1, 1, 225, s0*s1]" = torch.ops.prims.convert_element_type.default(div, torch.float16)
        unsqueeze: "f16[1, 1, 225, s0*s1, 1]" = torch.ops.aten.unsqueeze.default(convert_element_type_2, 4);  convert_element_type_2 = None
        permute: "f16[1, 1, s0*s1, 1, 225]" = torch.ops.aten.permute.default(unsqueeze, [0, 1, 3, 4, 2]);  unsqueeze = None
        permute_2: "f16[s0*s1, 225, 1, 1, 1]" = torch.ops.aten.permute.default(permute, [2, 4, 0, 1, 3]);  permute = None
        view_1: "f16[1, s0*s1, 225]" = torch.ops.aten.reshape.default(permute_2, [1, sym_size_int, 225]);  permute_2 = None
        convert_element_type_3: "f16[1, 256, 225]" = torch.ops.prims.convert_element_type.default(arg7_1, torch.float16);  arg7_1 = None
        unsqueeze_1: "f16[1, 256, 225, 1]" = torch.ops.aten.unsqueeze.default(convert_element_type_3, 3);  convert_element_type_3 = None
        unsqueeze_2: "f16[1, 256, 225, 1, 1]" = torch.ops.aten.unsqueeze.default(unsqueeze_1, 4);  unsqueeze_1 = None
        permute_1: "f16[1, 1, 1, 256, 225]" = torch.ops.aten.permute.default(unsqueeze_2, [3, 0, 4, 1, 2]);  unsqueeze_2 = None
        permute_3: "f16[225, 1, 1, 256, 1]" = torch.ops.aten.permute.default(permute_1, [4, 0, 1, 3, 2]);  permute_1 = None
        view_2: "f16[1, 225, 256]" = torch.ops.aten.reshape.default(permute_3, [1, 225, 256]);  permute_3 = None
        bmm: "f16[1, s0*s1, 256]" = torch.ops.aten.bmm.default(view_1, view_2);  view_2 = None
        view_3: "f16[s0*s1, 1, 1, 1, 256]" = torch.ops.aten.reshape.default(bmm, [sym_size_int, 1, 1, 1, 256]);  bmm = sym_size_int = None
        permute_4: "f16[1, 1, s0*s1, 256, 1]" = torch.ops.aten.permute.default(view_3, [2, 3, 0, 4, 1]);  view_3 = None
        sym_size_int_1: "Sym(s0*s1)" = torch.ops.aten.sym_size.int(view_1, 1);  view_1 = None
        view_4: "f16[1, 1, s0*s1, 256]" = torch.ops.aten.reshape.default(permute_4, [1, 1, sym_size_int_1, 256]);  permute_4 = sym_size_int_1 = None
        
        # File: /workspace/networks/layers/attention.py:378 in torch_dynamo_resume_in_forward_at_346, code: 3).reshape(h * w, n, c)
        add_3: "f16[1, 1, s10*s11, 256]" = torch.ops.aten.add.Tensor(view_9, view_4);  view_9 = view_4 = None
        permute_7: "f16[s10*s11, 1, 1, 256]" = torch.ops.aten.permute.default(add_3, [2, 0, 1, 3]);  add_3 = None
        
        # File: /workspace/networks/layers/attention.py:379 in torch_dynamo_resume_in_forward_at_346, code: else:
        view_10: "f16[s10*s11, 1, 256]" = torch.ops.aten.reshape.default(permute_7, [mul, 1, 256]);  permute_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)
        view_11: "f16[s10*s11, 256]" = torch.ops.aten.reshape.default(view_10, [mul, 256]);  view_10 = None
        convert_element_type_10: "f16[256, 256]" = torch.ops.prims.convert_element_type.default(arg16_1, torch.float16);  arg16_1 = None
        permute_8: "f16[256, 256]" = torch.ops.aten.permute.default(convert_element_type_10, [1, 0]);  convert_element_type_10 = None
        addmm: "f16[s10*s11, 256]" = torch.ops.aten.addmm.default(convert_element_type_9, view_11, permute_8);  convert_element_type_9 = view_11 = permute_8 = None
        view_12: "f16[s10*s11, 1, 256]" = torch.ops.aten.reshape.default(addmm, [mul, 1, 256]);  addmm = mul = None
        return (view_12, div)