github
/
ChatGLM-6B
mirror of https://github.com/THUDM/ChatGLM-6B
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Update loading instructions

pull/621/head
duzx16 2023-04-13 16:10:10 +08:00
parent 63b1e4e804
commit 5ade1e4055
2 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
ptuning/README.md
View File

@ -152,7 +152,8 @@ model = AutoModel.from_pretrained("THUDM/chatglm-6b", config=config, trust_remot
prefix_state_dict = torch.load(os.path.join(CHECKPOINT_PATH, "pytorch_model.bin"))
new_prefix_state_dict = {}
for k, v in prefix_state_dict.items():
new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
if k.startswith("transformer.prefix_encoder."):
new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
```
注意你可能需要将 `pre_seq_len` 改成你训练时的实际值。如果你是[从本地加载模型的话](https://github.com/THUDM/ChatGLM-6B#%E4%BB%8E%E6%9C%AC%E5%9C%B0%E5%8A%A0%E8%BD%BD%E6%A8%A1%E5%9E%8B)，需要将 `THUDM/chatglm-6b` 改成本地的模型路径（注意不是checkpoint路径）。
@ -160,7 +161,7 @@ model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
(2) 如果需要加载的是旧 Checkpoint（包含 ChatGLM-6B 以及 PrefixEncoder 参数），或者进行的全参数微调，则直接加载整个 Checkpoint：
```python
model = AutoModel.from_pretrained(CHECKPOINT_PATH, config=config, trust_remote_code=True)
model = AutoModel.from_pretrained(CHECKPOINT_PATH, trust_remote_code=True)
```
之后根据需求可以进行量化，也可以直接使用：

3
ptuning/main.py
View File

@ -118,7 +118,8 @@ def main():
prefix_state_dict = torch.load(os.path.join(model_args.ptuning_checkpoint, "pytorch_model.bin"))
new_prefix_state_dict = {}
for k, v in prefix_state_dict.items():
new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
if k.startswith("transformer.prefix_encoder."):
new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
else:
model = AutoModel.from_pretrained(model_args.model_name_or_path, config=config, trust_remote_code=True)