AkimfromParis / RAG-Japanese

Open source RAG with Llama Index for Japanese LLM in low resource settting

Open-source RAG for Japanese LLM in low-resource settings

1. Llama-Index as a data framework
2. Embeddings from Multi-lingual E5 from Microsoft
3. Elyza as LLM based on LlaMA 2 (Facebook/Meta)
4. 4Bits quantized by Bitsandbytes (served by Hugging Face Accelerate)
5. CJKPDFReader loader for Japanese PDF files
6. Open source Vector database (Postgres + PGVector)
7. Vanilla hybrid search
8. Ngrok for TCP tunneling
9. Q&A test
10. Conclusion