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

UNSLOTH_ENABLE_FULL_FINETUNING

Browse source
This commit is contained in:
Daniel Han 2025-03-09 22:57:24 -07:00
commit d41e5578e1
1 changed files with 8 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

9
unsloth/models/vision.py
View file

@ -172,10 +172,13 @@ class FastBaseModel:
pass
if full_finetuning:
os.environ["UNSLOTH_ENABLE_FULL_FINETUNING"] = "1"
if dtype == torch.bfloat16:
print("Unsloth: Using bfloat16 full finetuning which cuts memory usage by 50%.")
else:
print("Unsloth: Float16 full finetuning uses more memory since we upcast weights to float32.")
else:
os.environ["UNSLOTH_ENABLE_FULL_FINETUNING"] = "0"
pass
kwargs.pop("attn_implementation", None); # No need since we auto call it
@ -287,6 +290,10 @@ class FastBaseModel:
temporary_location = "_unsloth_temporary_saved_buffers",
**kwargs,
):
if os.environ.get("UNSLOTH_ENABLE_FULL_FINETUNING", "0") == "1":
print("Unsloth: Full finetuning is enabled, so .get_peft_model has no effect")
return model
pass
transformers_set_seed(random_state)
if type(r) is not int:
@ -363,7 +370,7 @@ class FastBaseModel:
"Unsloth: Your model needs to call `.get_peft_model` first!"
)
pass
full_finetuning = hasattr(model.config, "quantization_config", None) is not None
full_finetuning = os.environ.get("UNSLOTH_ENABLE_FULL_FINETUNING", "0") == "1"
float32_mixed_precision = True
if _get_dtype(model.config.torch_dtype) == torch.bfloat16: