423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-05 21:22:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix ORPO loss

Browse Source 
Former-commit-id: 816d71414617590f95de89a49f38358e597ed121
This commit is contained in:
hiyouga 2024-04-01 14:42:41 +08:00
parent 69e1d39832
commit bd52e2b404
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/llmtuner/train/orpo/trainer.py
View File

@ -99,7 +99,7 @@ class CustomORPOTrainer(DPOTrainer):
""" """
metrics = {} metrics = {}
chosen_logps, rejected_logps, chosen_logits, rejected_logits = self.concatenated_forward(model, batch) chosen_logps, rejected_logps, chosen_logits, rejected_logits = self.concatenated_forward(model, batch)
sft_loss = chosen_logps sft_loss = -chosen_logps
odds_ratio_loss = self.odds_ratio_loss(chosen_logps, rejected_logps) odds_ratio_loss = self.odds_ratio_loss(chosen_logps, rejected_logps)
batch_loss = (sft_loss + self.beta * odds_ratio_loss).mean() batch_loss = (sft_loss + self.beta * odds_ratio_loss).mean()