p04798526
/
LLaMA-Factory-Mirror
1
0
Fork 1
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #900 from mmbwf/main 

Fix PPO replace_model loading wrong default parameters.
This commit is contained in:
hoshi-hiyouga 2023-09-14 17:34:22 +08:00 committed by GitHub
parent bd4e24bfee 30fb721f12
commit 3202985087
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
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({