riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 23 Packages Projects Releases Wiki Activity

Merge pull request #1154 from ziliwang/master

Browse Source 
fix: hard coding for max number
This commit is contained in:
Thomas Wolf 2019-08-30 23:23:08 +02:00 committed by GitHub
commit 206c35e9a4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
pytorch_transformers/modeling_xlnet.py
View File

@ -418,7 +418,10 @@ class XLNetRelativeAttention(nn.Module):
attn_score = (ac + bd + ef) * self.scale
if attn_mask is not None:
# attn_score = attn_score * (1 - attn_mask) - 1e30 * attn_mask
attn_score = attn_score - 1e30 * attn_mask
if attn_mask.dtype == torch.float16:
attn_score = attn_score - 65500 * attn_mask
else:
attn_score = attn_score - 1e30 * attn_mask
# attention probability
attn_prob = F.softmax(attn_score, dim=1)