ChatGLM-6B icon indicating copy to clipboard operation
ChatGLM-6B copied to clipboard

通过ptuning微调参数加载checkpoint后,推理逻辑不对了

Open guarx opened this issue 1 year ago • 5 comments

Is there an existing issue for this?

  • [X] I have searched the existing issues

Current Behavior

image 这个是不是有问题呢, 加载训练完后的checkpoint后,之前的推理逻辑都没了,这是加载训练checkpoint的代码 image 这个结果明显不对啊 image 训练用的json就是文档里的

Expected Behavior

No response

Steps To Reproduce

  1. bash ptuning/train.sh
  2. bash ptuning/evaluate.sh
  3. streamlit run web_demo2.py --server.port 6006

Environment

- OS: ubuntu
- Python: 3.8
- Transformers: 4.27
- PyTorch:
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) : True

Anything else?

No response

guarx avatar Apr 20 '23 06:04 guarx

同样的问题。

感觉这个模型根本不能调

liaoweiguo avatar Apr 20 '23 06:04 liaoweiguo

同样的问题

derekzhuo avatar Apr 20 '23 08:04 derekzhuo

我的是微调完后只能回答喵喵了

waveboy800 avatar Apr 25 '23 07:04 waveboy800

我的是微调完后只能回答喵喵了

你好,问一下哈,你的微调训练参数有自己调整吗?我的微调完推理时答复是空的,并且推理速度特别慢

sun1092469590 avatar May 05 '23 07:05 sun1092469590

没有

waveboy800 avatar May 05 '23 08:05 waveboy800

我也遇到了,ptuning v2调过之后就变傻子了。

xmlv avatar Jun 11 '23 10:06 xmlv

Duplicate of #403

zhangch9 avatar Aug 16 '23 11:08 zhangch9