class GraphModule(torch.nn.Module):
    def forward(self, s0: "Sym(s0)", s1: "Sym(s1)", L_v_: "f16[1, 256, s0, s1][256*s0*s1, s0*s1, s1, 1]cuda:0", L_self_modules_relative_emb_k_parameters_weight_: "f32[225, 256, 1, 1][256, 1, 1, 1]cuda:0", L_self_modules_relative_emb_k_parameters_bias_: "f32[225][1]cuda:0", L_q_: "f16[1, 256, s0, s1][256*s0*s1, s0*s1, s1, 1]cuda:0", L_k_: "f16[1, 256, s0, s1][256*s0*s1, s0*s1, s1, 1]cuda:0"):
        l_v_ = L_v_
        l_self_modules_relative_emb_k_parameters_weight_ = L_self_modules_relative_emb_k_parameters_weight_
        l_self_modules_relative_emb_k_parameters_bias_ = L_self_modules_relative_emb_k_parameters_bias_
        l_q_ = L_q_
        l_k_ = L_k_
        
        # File: /workspace/networks/layers/attention.py:316 in forward, code: n, c, h, w = v.size()
        size = l_v_.size()
        getitem_2: "Sym(s0)" = size[2]
        getitem_3: "Sym(s1)" = size[3];  size = None
        
        # File: /workspace/networks/layers/attention.py:327 in forward, code: memory_mask = torch.ones((1, 1, h, w), device=v.device).float()
        ones: "f32[1, 1, s0, s1][s0*s1, s0*s1, s1, 1]cuda:0" = torch.ones((1, 1, getitem_2, getitem_3), device = device(type='cuda', index=0))
        memory_mask: "f32[1, 1, s0, s1][s0*s1, s0*s1, s1, 1]cuda:0" = ones.float();  ones = 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: "f32[1, 1, s0 + 14, s1 + 14][s0*s1 + 14*s0 + 14*s1 + 196, s0*s1 + 14*s0 + 14*s1 + 196, s1 + 14, 1]cuda:0" = torch._C._nn.pad(memory_mask, (7, 7, 7, 7), 'constant', 0);  memory_mask = None
        
        # File: /workspace/networks/layers/attention.py:434 in pad_and_unfold, code: x = F.unfold(x,
        x_1: "f32[1, 225, s0*s1][225*s0*s1, s0*s1, 1]cuda:0" = torch.nn.functional.unfold(x, kernel_size = (15, 15), stride = (1, 1), dilation = 1);  x = None
        
        # File: /workspace/networks/layers/attention.py:329 in forward, code: 1, 1, self.window_size * self.window_size, h * w)
        mul: "Sym(s0*s1)" = getitem_2 * getitem_3
        
        # File: /workspace/networks/layers/attention.py:328 in forward, code: unfolded_k_mask = self.pad_and_unfold(memory_mask).view(
        unfolded_k_mask: "f32[1, 1, 225, s0*s1][225*s0*s1, 225*s0*s1, s0*s1, 1]cuda:0" = x_1.view(1, 1, 225, mul);  x_1 = mul = None
        
        # File: /workspace/networks/layers/attention.py:330 in forward, code: qk_mask = 1 - unfolded_k_mask
        qk_mask: "f32[1, 1, 225, s0*s1][225*s0*s1, 225*s0*s1, s0*s1, 1]cuda:0" = 1 - unfolded_k_mask;  unfolded_k_mask = None
        
        # 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,
        relative_emb: "f16[1, 225, s0, s1][225*s0*s1, s0*s1, s1, 1]cuda:0" = torch.conv2d(l_q_, l_self_modules_relative_emb_k_parameters_weight_, l_self_modules_relative_emb_k_parameters_bias_, (1, 1), (0, 0), (1, 1), 1);  l_self_modules_relative_emb_k_parameters_weight_ = l_self_modules_relative_emb_k_parameters_bias_ = None
        
        # File: /workspace/networks/layers/attention.py:335 in forward, code: q = q / self.T
        q: "f16[1, 256, s0, s1][256*s0*s1, s0*s1, s1, 1]cuda:0" = l_q_ / 16.0;  l_q_ = None
        
        # File: /workspace/networks/layers/attention.py:337 in forward, code: q = q.view(-1, self.d_att, h, w)
        q_1: "f16[1, 256, s0, s1][256*s0*s1, s0*s1, s1, 1]cuda:0" = q.view(-1, 256, getitem_2, getitem_3);  q = None
        
        # File: /workspace/networks/layers/attention.py:338 in forward, code: k = k.view(-1, self.d_att, h, w)
        k: "f16[1, 256, s0, s1][256*s0*s1, s0*s1, s1, 1]cuda:0" = l_k_.view(-1, 256, getitem_2, getitem_3);  l_k_ = None
        
        # File: /workspace/networks/layers/attention.py:339 in forward, code: v = v.view(-1, self.num_head, hidden_dim, h * w)
        mul_1: "Sym(s0*s1)" = getitem_2 * getitem_3
        v: "f16[1, 1, 256, s0*s1][256*s0*s1, 256*s0*s1, s0*s1, 1]cuda:0" = l_v_.view(-1, 1, 256, mul_1);  l_v_ = mul_1 = None
        
        # File: /workspace/networks/layers/attention.py:343 in forward, code: h * w)
        mul_2: "Sym(s0*s1)" = getitem_2 * getitem_3
        
        # File: /workspace/networks/layers/attention.py:341 in forward, code: relative_emb = relative_emb.view(n, self.num_head,
        relative_emb_1: "f16[1, 1, 225, s0*s1][225*s0*s1, 225*s0*s1, s0*s1, 1]cuda:0" = relative_emb.view(1, 1, 225, mul_2);  relative_emb = mul_2 = None
        return (q_1, k, v, getitem_2, getitem_3, qk_mask, relative_emb_1)