HiRAG
HiRAG copied to clipboard
[EMNLP'25 findings] This is the official repo for the paper, HiRAG: Retrieval-Augmented Generation with Hierarchical Knowledge.
Consistently see warning in console logs when processing a single file (academic pdf as markdown): RuntimeWarning: invalid value encountered in scalar divide if (abs(prev_bic - bic) / abs(prev_bic)) < rel_tol:...
pip struggles finding Langchain packages needed by ragas. processes halts: --- Using cached langchain_community-0.0.14-py3-none-any.whl.metadata (7.5 kB) Using cached langchain_community-0.0.13-py3-none-any.whl.metadata (7.5 kB) Using cached langchain_community-0.0.12-py3-none-any.whl.metadata (7.5 kB) Using cached langchain_community-0.0.11-py3-none-any.whl.metadata (7.3...
Hi, I noticed that you repeated the description of the goal here. Does this provide any performance gain? If so, roughly how much improvement does it offer? https://github.com/hhy-huang/HiRAG/blob/aaedbf955bfcf05925d9ce21ec50d3014d68b37c/hirag/prompt.py#L548-L585
Hi, Very interesting work ! How can we run it with Ollama to access to open LLM and with Hugging Face for embeddings ? Thank you.
Smalll adjustments to the codebase to work with ollama, cohere and deepseek. Also created a script repo https://github.com/georgiedekker/hi_rag_ollama_cohere_scripts.git