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

Update mistral.py, showed flag to not call cut cross entropy (#3233)

Browse source 
* Update mistral.py, showed flag to not call cut cross entropy

* Update mistral.py, made it so if its not equal to zero

* Update unsloth/models/mistral.py

---------

Co-authored-by: Daniel Han <danielhanchen@gmail.com>
This commit is contained in:
pluesclues 2025-08-29 04:32:21 -04:00 • committed by GitHub
commit fb4ae0134a
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
unsloth/models/mistral.py
View file

@ -300,7 +300,7 @@ def MistralForCausalLM_fast_forward(
# < 1024 Normal Unsloth uses less VRAM!
if bsz * q_len <= 1024: RETURN_LOGITS = True
if not RETURN_LOGITS and HAS_CUT_CROSS_ENTROPY and labels is not None:
if not RETURN_LOGITS and HAS_CUT_CROSS_ENTROPY and os.environ.get("UNSLOTH_ENABLE_CCE", "1") != "0" and labels is not None:
n_items = kwargs.get("num_items_in_batch", None) or kwargs.get("n_items", None)
logit_softcapping = getattr(self.config, "final_logit_softcapping", 0)
loss = fused_linear_cross_entropy(