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

fix: compute_loss bug (#1151)

Browse source 
Currently, Unsloth doesn't pass additional parameters to Trainer.compute_loss such as return_outputs. This leads to errors when calling trainer.evaluate(). This change fixes the bug by properly passing parameters to Trainer.compute_loss.
This commit is contained in:
vo1d-ai 2024-10-18 23:46:07 -04:00 • committed by GitHub
commit 5dcc0cd613
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
unsloth/models/_utils.py
View file

@ -1166,7 +1166,7 @@ def _unsloth_pre_compute_loss(self, model, inputs, *args, **kwargs):
inputs["num_items_in_batch"] = kwargs["num_items_in_batch"]
pass
pass
return self._old_compute_loss(model, inputs, args, kwargs)
return self._old_compute_loss(model, inputs, *args, **kwargs)
pass