423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-27 08:12:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #900 from mmbwf/main

Browse Source 
Fix PPO replace_model loading wrong default parameters.

Former-commit-id: 3202985087870f65d639d5f0891ad90d7b39f2ef
This commit is contained in:
hoshi-hiyouga 2023-09-14 17:34:22 +08:00 committed by GitHub
commit fe2ac71d11
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llmtuner/tuner/ppo/utils.py
View File

@ -10,8 +10,8 @@ if TYPE_CHECKING:
def replace_model(model: "AutoModelForCausalLMWithValueHead", target: Literal["default", "reward"]) -> None:
if target == "reward": # save default head temporarily
valuehead_state_dict = model.v_head.state_dict()
setattr(model, "default_head_weight", valuehead_state_dict["summary.weight"])
setattr(model, "default_head_bias", valuehead_state_dict["summary.bias"])
setattr(model, "default_head_weight", valuehead_state_dict["summary.weight"].clone())
setattr(model, "default_head_bias", valuehead_state_dict["summary.bias"].clone())
model.pretrained_model.set_adapter(target) # set the LoRA adapter to be active
model.v_head.load_state_dict({