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

Sync UNSLOTH_FORCE_FLOAT32 in forced float32 trainer path

Browse source 
(cherry picked from commit d27c104ec831e8e00571206945cafbc31308fcf1)
This commit is contained in:
Daniel Han 2026-02-25 16:40:50 +00:00
commit feabba825a
1 changed files with 1 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
unsloth/models/rl.py
View file

@ -733,6 +733,7 @@ def _patch_trl_rl_trainers(trainer_file = "grpo_trainer"):
" force_float32 = True\n"
"if force_float32:\n"
" # Forced float32 training\n"
" os.environ['UNSLOTH_FORCE_FLOAT32'] = '1'\n"
" args.fp16 = False\n"
" args.bf16 = False\n"
" os.environ['ACCELERATE_MIXED_PRECISION'] = 'no'\n"