Qdrant experiments: 10→500 candidate depth lifts best-possible nDCG by 0.28 but real score by ≤0.01
qdrant_engine · x · 2026-09-21
Qdrant tested candidate depth across five public datasets: raising the limit from 10 to 500 improved the best-possible nDCG@10 by up to 0.28, yet the actual pipeline score changed by at most 0.01—documents are often retrieved, just not ranked highly enough. Practical guidance: start testing at limit 100 and 200 for reranker stages, compare approximate recall against exact search before raising hnswef, and try quantization before cutting depth when RAM-bound.
More from coding & agent
- Gradio Workflow launches on Product Hunt: node-based AI pipelines powered by Hugging Face — Gradio · 2026-09-21
- AGNTcon/MCPcon Amsterdam Draws Excited Crowd Around MCP's Future — dsp_ · 2026-09-21
- AWS launches serverless AgentCore Runtime as agents go always-on — bookwormengr · 2026-09-21
- Dev Uses Jev to Dynamically Tune GPT-6 Reasoning Effort Inside Codex, Cutting Costs 50% — brandon_galang · 2026-09-21
- Question-first prompting took local Qwen from 89% to 100% and cut latency 5x — Micha0827 · 2026-09-21
- Open-source auto-router prices cache-aware model routing for Claude Code and Cursor — airesearch12 · 2026-09-21