Papers with Code Search Engine: Powered by Hugging Face Infrastructure
NielsRogge · x · 2026-08-25
Niels Rogge from Hugging Face details the tech stack behind the Papers with Code search engine, designed to help humans and agents find relevant research.
Technical Details:
- Hybrid Search: Combines keyword and vector search for superior results.
- Embedding Model: Uses Alibaba's Qwen3-Embedding-0.6B.
- Infrastructure: PostgreSQL with pgvector; embeddings computed on NVIDIA L4 GPUs via Hugging Face Jobs; artifacts stored in Buckets.
- Inference: Live embedding endpoints served via Hugging Face Inference Endpoints, powering the "related papers" feature.
Related event: Hugging Face Details How Papers with Code Search Is Built(2 posts)→
More from Infra
- DevOps to AI Infra is becoming a serious career path — _jaydeepkarale · 2026-08-25
- AI Growth Forces Smarter Cloud: Power Becomes the New Bottleneck — DavidLinthicum · 2026-08-25
- Nvidia Claims Groq 3 LPX 4x Faster Than Cerebras, But Needs 64 Accelerators — The Decoder · 2026-08-25
- Is Agent Collaboration the Next Major AI Infrastructure Layer? — Plenty-Ad-8268 · 2026-08-25
- Goldman: China's advanced chip self-sufficiency to reach 66% by 2035 — pstAsiatech · 2026-08-25
- Quantization-Aware Healing: Recovering 4-Bit LLMs Faster — MultiverseComputingCAI · 2026-08-25