ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update llama.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-03-18 03:30:00 +11:00
commit b283a6bf08
1 changed files with 6 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

10
unsloth/models/llama.py
View file

@ -1385,14 +1385,16 @@ class FastLlamaModel:
# Now patch lm_head and embed_tokens
if train_embed_tokens:
print("Unsloth: Casting embed_tokens to float32")
model.model.model.embed_tokens.to(torch.float32)
model.model.model.embed_tokens.requires_grad_(True)
assert(hasattr(model.model.model.embed_tokens, "modules_to_save"))
model.model.model.embed_tokens.modules_to_save.default.to(torch.float32)
model.model.model.embed_tokens.modules_to_save.default.requires_grad_(True)
pass
if train_lm_head:
print("Unsloth: Casting lm_head to float32")
model.model.lm_head.to(torch.float32)
model.model.lm_head.requires_grad_(True)
assert(hasattr(model.model.lm_head, "modules_to_save"))
model.model.lm_head.modules_to_save.default.to(torch.float32)
model.model.lm_head.modules_to_save.default.requires_grad_(True)
pass
return model