ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix #4742

Browse Source 
Former-commit-id: ae9cf84347878fcc462f35db941c14e1df104276
This commit is contained in:
hiyouga
2024-07-09 23:24:24 +08:00
parent 11d961cf3c
commit 834c4e8ad9
2 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/llamafactory/train/dpo/trainer.py
View File

@@ -54,6 +54,7 @@ class CustomDPOTrainer(DPOTrainer):
disable_dropout_in_model(ref_model) disable_dropout_in_model(ref_model)
self.finetuning_args = finetuning_args self.finetuning_args = finetuning_args
self.f_divergence_type = "reverse_kl"
self.reference_free = False self.reference_free = False
self.use_dpo_data_collator = True # hack to avoid warning self.use_dpo_data_collator = True # hack to avoid warning
self.generate_during_eval = False # disable at evaluation self.generate_during_eval = False # disable at evaluation