[Feature] Locale Support

[Ucag]

What feature would you like to see?

Thanks for your great work. Here is a feature I think is very useful.

When instructing prompt model to generate candidates, the language is in English. However, for many other use cases, many tasks may be language-related, like writing poem or lyrics. This requires the generated instruction to be the specific language. What's more, like CoT in deepseek, the CoT process is in Chinese ( for best CoT support, according to my use experience).

In my case, my task is a classification task in non-English language. I found out that the LLM could output right result, however the result is in English. The problem disappears after I manually translate the trained instruction into target language.

I just intuitionally draw a sketch is that DSPy can do a simple translation using LLM during training or a post-process instruction class. I'm not sure whether this method is robust enough to cover all locale-related tasks.

Do you have any plan for this?

Would you like to contribute?

• Yes, I'd like to help implement this.
• No, I just want to request it.

Additional Context

No response

[chenmoneygithub]

@Ucag Thanks for reporting the issue! Yes locale support is something we are looking into, we cannot promise an exact ETA now, but should come no later than summer.