Source
How Hugging Face Inference Endpoints, Jobs, and Buckets Power Search on Papers with Code
Hugging Face developed a hybrid search system for Papers with Code that combines lexical and vector-based retrieval. The architecture utilizes Hugging Face Jobs, Storage Buckets, and Inference Endpoints to power the engine.
Why it matters
Researchers can find relevant AI papers using natural language queries instead of just exact keywords. The system remains functional via a lexical fallback if semantic services are temporarily unavailable.
The details
- Employs Qwen/Qwen3-Embedding-0.6B to generate 256-dimensional vectors.
- Reciprocal Rank Fusion (RRF) combines results from lexical and semantic search.
- Maintains embeddings for over 110,000 papers from arXiv and Daily Papers.
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.