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

Fix: skip fp16/bf16 validation for full finetuning in RL trainers (#6813)

Browse source 
---------

Co-authored-by: Ayushman Paul <ayushman@HP>
This commit is contained in:
Ayushman 2026-07-03 00:11:10 +05:30 • committed by GitHub
commit d33a7a7a1a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 3 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
unsloth/models/rl.py
View file

@ -1015,6 +1015,9 @@ def _patch_trl_rl_trainers_impl(trainer_file = "grpo_trainer"):
"dtype = _get_dtype(dtype)\n"
"float16 = dtype == torch.float16\n"
"bfloat16 = dtype == torch.bfloat16\n"
"if full_finetuning:\n"
" if bfloat16 and use_fp16: use_fp16 = False\n"
" if float16 and use_bf16: use_bf16 = False\n"
"if not force_float32 and (float16 and use_bf16): raise TypeError('Unsloth: Model is in float16 precision but you want to use bfloat16 precision. Set fp16 to `True` and bf16 to `False`')\n"
"if not force_float32 and (bfloat16 and use_fp16): raise TypeError('Unsloth: Model is in bfloat16 precision but you want to use float16 precision. Set fp16 to `False` and bf16 to `True`')\n"
"if force_float32:\n"