mirror of
https://github.com/osmarks/nanogpt-experiments.git
synced 2024-12-18 06:00:29 +00:00
Fix GPT.crop_block_size when flash attention is available
This commit is contained in:
parent
a82b33b525
commit
c3f254844d
3
model.py
3
model.py
@ -207,7 +207,8 @@ class GPT(nn.Module):
|
||||
self.config.block_size = block_size
|
||||
self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size])
|
||||
for block in self.transformer.h:
|
||||
block.attn.bias = block.attn.bias[:,:,:block_size,:block_size]
|
||||
if hasattr(block.attn, 'bias'):
|
||||
block.attn.bias = block.attn.bias[:,:,:block_size,:block_size]
|
||||
|
||||
@classmethod
|
||||
def from_pretrained(cls, model_type, override_args=None):
|
||||
|
Loading…
Reference in New Issue
Block a user