423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-12 18:08:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix layer norm name in PPO

Browse Source 
Former-commit-id: e3aaef7d4a37e4aa388a9158c382db8239843a5e
This commit is contained in:
hiyouga 2023-06-02 17:30:01 +08:00
parent b0e9a673be
commit 6ab22a0181
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/utils/ppo.py
View File

@ -41,7 +41,7 @@ def replace_model(model: AutoModelForCausalLMWithValueHead, target: Literal["def
def cast_layernorm_dtype( def cast_layernorm_dtype(
model: AutoModelForCausalLMWithValueHead, model: AutoModelForCausalLMWithValueHead,
layer_norm_names: List[str] = ["layernorm"], # for chatglm setting layer_norm_names: List[str] = ["norm", "ln_f"], # for LLaMA and BLOOM setting
layer_norm_params: Optional[Dict[str, torch.Tensor]] = None layer_norm_params: Optional[Dict[str, torch.Tensor]] = None
) -> Tuple[AutoModelForCausalLMWithValueHead, Dict[str, torch.Tensor]]: ) -> Tuple[AutoModelForCausalLMWithValueHead, Dict[str, torch.Tensor]]: