github
/
minimind
mirror of https://github.com/jingyaogong/minimind.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix: Loading LoRA parameters which saved from multi-card training

pull/523/head
yuyu5333 2025-11-06 19:03:18 +08:00
parent 81e869fc3e
commit 7de4b9ecd6
2 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@ -1,4 +1,5 @@
model/__pycache__ model/__pycache__
trainer/__pycache__
out out
website/ website/
docs-minimind/ docs-minimind/

4
model/model_lora.py
View File

@ -34,9 +34,11 @@ def apply_lora(model, rank=8):
def load_lora(model, path): def load_lora(model, path):
state_dict = torch.load(path, map_location=model.device) state_dict = torch.load(path, map_location=model.device)
# 去除权重键名中可能存在的 module. 前缀
cleaned_state_dict = {k.replace('module.', ''): v for k, v in state_dict.items()}
for name, module in model.named_modules(): for name, module in model.named_modules():
if hasattr(module, 'lora'): if hasattr(module, 'lora'):
lora_state = {k.replace(f'{name}.lora.', ''): v for k, v in state_dict.items() if f'{name}.lora.' in k} lora_state = {k.replace(f'{name}.lora.', ''): v for k, v in cleaned_state_dict.items() if f'{name}.lora.' in k}
module.lora.load_state_dict(lora_state) module.lora.load_state_dict(lora_state)