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

update ppo trainer

Browse Source 
Former-commit-id: caa525a5c6f228b9ad71387d1fe4f1c2ffa2479e
This commit is contained in:
hiyouga
2023-11-20 21:39:15 +08:00
parent e585950c54
commit 28258aecd2
7 changed files with 68 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llmtuner/model/loader.py
View File

@@ -202,7 +202,7 @@ def load_model_and_tokenizer(
if stage in ["rm", "ppo"]:
model: "AutoModelForCausalLMWithValueHead" = AutoModelForCausalLMWithValueHead.from_pretrained(model)
setattr(model, "_keys_to_ignore_on_save", [name for name, _ in model.named_parameters() if "pretrained_model" in name])
setattr(model, "tie_weights", MethodType(lambda _: None, model))
setattr(model, "tie_weights", MethodType(lambda _: None, model)) # use empty method
vhead_path = (
model_args.checkpoint_dir[-1] if model_args.checkpoint_dir is not None else model_args.model_name_or_path
)