mirror of
https://github.com/huggingface/transformers.git
synced 2025-08-01 02:31:11 +06:00
Cloned tensors after indexing in _compute_attn_output_with_global_indices (#13613)
Co-authored-by: Alessandro Suglia <asuglia@fb.com>
This commit is contained in:
parent
ce32c69c0b
commit
19b7acdd61
@ -586,7 +586,7 @@ class LEDEncoderSelfAttention(nn.Module):
|
|||||||
# attn = torch.einsum('blhs,bshd->blhd', (selected_attn_probs, selected_v))
|
# attn = torch.einsum('blhs,bshd->blhd', (selected_attn_probs, selected_v))
|
||||||
# compute attn output only global
|
# compute attn output only global
|
||||||
attn_output_only_global = torch.matmul(
|
attn_output_only_global = torch.matmul(
|
||||||
attn_probs_only_global.transpose(1, 2), value_vectors_only_global.transpose(1, 2)
|
attn_probs_only_global.transpose(1, 2).clone(), value_vectors_only_global.transpose(1, 2).clone()
|
||||||
).transpose(1, 2)
|
).transpose(1, 2)
|
||||||
|
|
||||||
# reshape attn probs
|
# reshape attn probs
|
||||||
|
@ -976,7 +976,7 @@ class LongformerSelfAttention(nn.Module):
|
|||||||
# attn = torch.einsum('blhs,bshd->blhd', (selected_attn_probs, selected_v))
|
# attn = torch.einsum('blhs,bshd->blhd', (selected_attn_probs, selected_v))
|
||||||
# compute attn output only global
|
# compute attn output only global
|
||||||
attn_output_only_global = torch.matmul(
|
attn_output_only_global = torch.matmul(
|
||||||
attn_probs_only_global.transpose(1, 2), value_vectors_only_global.transpose(1, 2)
|
attn_probs_only_global.transpose(1, 2).clone(), value_vectors_only_global.transpose(1, 2).clone()
|
||||||
).transpose(1, 2)
|
).transpose(1, 2)
|
||||||
|
|
||||||
# reshape attn probs
|
# reshape attn probs
|
||||||
|
Loading…
Reference in New Issue
Block a user