Sql generated with additional Chinese

[njalan]

Sometime there are Chinese at the end of Sql :

[zainhoda]

@njalan could you provide some additional context here on the setup (LLM, training data, etc)

[njalan]

@zainhoda I am using Baichuan-7B as LLM and training consists of querion-sql pairs and documentation for business knowledge(it is mixed with Chinese and englist.) All the question are asked by Chinese

[njalan]

Sometimes it provided me the same two query splitted by -- 或者(it means OR in Englist) So is any any prompt to avoid it?

[zainhoda]

I think the easiest path might be to override the extract_sql method:

https://github.com/vanna-ai/vanna/blob/main/src/vanna/base/base.py#L126-L150

You can provide your own extract_sql method that will remove the unnecessary characters

[njalan]

@zainhoda Can you please help to fix it? I am not familiar with Python/

[zyclove]

@zainhoda Can you support this feature? Looking forward to releasing it

[zainhoda]

What issue are you experiencing? Are you using the same LLM as the other user?

[zyclove]

@zainhoda We are developers in China. We hope that the input description is in Chinese and the think prompt output is also in Chinese. It would be better to support multiple languages.

[Mioooooo]

@zyclove Replace LLM and use qwen:32b or llama3:70b

[njalan]

When I use codeQwen1.5 and then didn't face this issue any more by now