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

Update vision.py

Browse source
This commit is contained in:
Daniel Han 2025-10-30 06:23:28 -07:00
commit be1c2ca95c
1 changed files with 4 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

4
unsloth/models/vision.py
View file

@ -156,6 +156,9 @@ def unsloth_base_fast_generate(
FastBaseModel.for_inference(self)
dtype = _get_dtype(dtype_from_config(self.config))
# Handle float32 cases
if os.environ.get("UNSLOTH_BFLOAT16_MIXED_PRECISION", "0") == "1":
dtype = torch.bfloat16
# Check if VLM
is_vlm = any(
@ -516,6 +519,7 @@ class FastBaseModel:
f"Unsloth: Using bfloat16 full finetuning which cuts memory usage by 50%.\n"
f"To enable float32 training, use `float32_mixed_precision = True` during FastLanguageModel.from_pretrained"
)
os.environ["UNSLOTH_BFLOAT16_MIXED_PRECISION"] = "1"
else:
print("Unsloth: Float16 full finetuning uses more memory since we upcast weights to float32.")
else: