osmarks/nanogpt-experiments
1
0
Fork 0
mirror of https://github.com/osmarks/nanogpt-experiments.git synced 2025-10-31 15:23:01 +00:00
Code Issues Releases Wiki Activity

Merge pull request #220 from python273/patch-1

Browse Source 
Fix GPT.crop_block_size when flash attention is available
This commit is contained in:
Andrej
2023-04-12 22:13:01 -07:00
committed by GitHub
parent 8aeea6d970 c3f254844d
commit ea24604b29
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
model.py
View File

@@ -207,6 +207,7 @@ class GPT(nn.Module):
self.config.block_size = block_size self.config.block_size = block_size
self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size]) self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size])
for block in self.transformer.h: for block in self.transformer.h:
if hasattr(block.attn, 'bias'):
block.attn.bias = block.attn.bias[:,:,:block_size,:block_size] block.attn.bias = block.attn.bias[:,:,:block_size,:block_size]
@classmethod @classmethod