Some excellent large models perform well in English conversation scenarios, such as Gemini-exp-1206, which has reached the level of OpenAI's O1 model in some aspects, but the Language Average score is very low. For non-native English speakers (Japanese, Chinese, French, Spanish, etc.), to achieve the best conversational effect and and does not affect the original conversation context.they need to manually translate their questions into English and then have the LLM's English responses translated back into their native language, which creates a sense of disconnection during continuous conversations.

I hope there is such a function:

In the front-end conversation interface, users are always in a state of conversing with the LLM in their native language.

But in the back-end: first, translate the user's question into English, then pass the English version of the question to the conversation model. Next, after the LLM returns the English version of the answer, translate it into the user's native language.

What the user needs to do: decide whether to enable this function and configure the LLM used for translation.