layer issu

layers issues

> /content/trans-vsumm/seq_sum.py(78)forward()
-> src = layer(src, src_mask)
(Pdb) src.shape
torch.Size([461, 1024, 256])
(Pdb) src_mask.shape
torch.Size([461, 1, 1, 1024])


> /content/trans-vsumm/seq_sum.py(78)forward()
-> src = layer(src, src_mask)
(Pdb) src.shape
torch.Size([461, 1024, 256])
(Pdb) src_mask.shape
torch.Size([461, 1, 1, 1024])



(Pdb) self.layers
ModuleList(
  (0): EncoderLayer(
    (self_attn_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
    (ff_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
    (self_attention): MultiHeadAttentionLayer(
      (fc_q): Linear(in_features=1024, out_features=1024, bias=False)
      (fc_k): Linear(in_features=1024, out_features=1024, bias=False)
      (fc_v): Linear(in_features=1024, out_features=1024, bias=False)
      (fc_o): Linear(in_features=1024, out_features=1024, bias=False)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (dropout): Dropout(p=0.1, inplace=False)
  )
)
(Pdb) n

(Pdb) n

RuntimeError: mat1 dim 1 must match mat2 dim 0
> /content/trans-vsumm/seq_sum.py(78)forward()
-> src = layer(src, src_mask)