yacud1/

arxiv-bge-base-818k

Pre-built vector search index over 817,961 arXiv papers (titles + abstracts), embedded with BAAI/bge-base-en-v1.5 into 768-dim vectors and packaged as a ready-to-query FAISS index. Why this exists: embedding ~818k abstracts through BGE-base is the slow, hardware-hungry step. Download this once and you skip it entirely — load the index, embed only your query string with the same model, and search. No re-vectorizing… See the full description on the dataset page: https://huggingface.co/datasets/yacud1/arxiv-bge-base-818k.