About language

[mc112611]

what language does GraphRAG support? Can I use Japanese document to do RAG？

[eyast]

I believe so. GraphRAG makes calls to an OpenAI or AzureOpenAI LLM, and they support Japanese in their tokenizers. Keep in mind that the instructions (and examples) provided to the LLM to do entity extraction and summarization are all in English. You might want to modify these prompts.

For example, when you intiate the indexer, you will see prompts in prompts\entity_extraction.txt. Line 20 says "Return output in English". You might want to modify this so that the result is returned in Japanese.

[mc112611]

I believe so. GraphRAG makes calls to an OpenAI or AzureOpenAI LLM, and they support Japanese in their tokenizers. Keep in mind that the instructions (and examples) provided to the LLM to do entity extraction and summarization are all in English. You might want to modify these prompts.

For example, when you intiate the indexer, you will see prompts in prompts\entity_extraction.txt. Line 20 says "Return output in English". You might want to modify this so that the result is returned in Japanese.

Thank you so much ， hope this project will support local LLM， not only chatGPT。

[jgbradley1]

Thank you so much ， hope this project will support local LLM， not only chatGPT。

We welcome contributions from the community 😉 .