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

[BUG/Help] <title>8-3-5+4=?这样的逻辑问题该怎么训练

Open rikeyz opened this issue 1 year ago • 2 comments

Is there an existing issue for this?

  • [X] I have searched the existing issues

Current Behavior

image 结果居然等于3,还一本正经地推理了一遍。

Expected Behavior

4

Steps To Reproduce

3090显卡下,启动不带量化的模型,提问 8-3-5+4=?,回答结果为3

Environment

- OS:Centos 7.9
- Python: python 3.8
- Transformers:
- PyTorch:
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :

Anything else?

No response

rikeyz avatar Apr 21 '23 16:04 rikeyz

忘记是不是这个模型了,查找了很多7b左右的模型,记得有大佬说这个模型容量下做逻辑推理(数学计算)是表显比较差且不好训练的,个人认为试试更大的模型或者等创作者大佬来看看

SailNow avatar Apr 23 '23 08:04 SailNow

我认为 它根本没有学会做加减法,只是记到了一些式子的答案而已。。。

cywjava avatar May 05 '23 03:05 cywjava

调用api呗?比如eval()

JasonChenJC avatar Jun 14 '23 02:06 JasonChenJC

Duplicate of #712

zhangch9 avatar Aug 16 '23 11:08 zhangch9