SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix missed head transpose

Browse Source
This commit is contained in:
Simon Layton
2019-10-03 09:23:16 -04:00
parent d51b589404
commit 9ffda216ec
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
transformers/modeling_xlnet.py
View File

@@ -284,7 +284,7 @@ class XLNetRelativeAttention(nn.Module):
# Mask heads if we want to # Mask heads if we want to
if head_mask is not None: if head_mask is not None:
attn_prob = attn_prob * head_mask attn_prob = attn_prob * torch.einsum('ijbn->bnij', head_mask)
# attention output # attention output
attn_vec = torch.einsum('bnij,jbnd->ibnd', attn_prob, v_head_h) attn_vec = torch.einsum('bnij,jbnd->ibnd', attn_prob, v_head_h)