423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-10-14 23:58:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix PPO trainer

Browse Source 
Former-commit-id: 21982a7d4dd9b7c3a1145b481f02b9990e32dc00
This commit is contained in:
hiyouga 2023-08-02 19:10:23 +08:00
parent e4d0b8ee6e
commit 8bd1da7144
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/tuner/ppo/trainer.py
View File

@ -161,7 +161,7 @@ class PPOPeftTrainer(PPOTrainer, PeftTrainer):
unwrapped_model.pretrained_model.generation_config._from_model_config = False
queries, responses = [], []
query, response = inputs["input_ids"], response[:, inputs["input_ids"].size(-1):].detach().cpu()
query, response = inputs["input_ids"].detach().cpu(), response[:, inputs["input_ids"].size(-1):].detach().cpu()
for i in range(len(query)):
query_length = (query[i] != self.tokenizer.pad_token_id).nonzero()[0]
response_length = (response[i] != self.tokenizer.pad_token_id).nonzero()[-1] + 1