Open sourcing parsing pipeline: 118 pages/min on a single H200
YTjay_SC · reddit · 2026-09-02
The author open-sourced 'Meridian,' a high-performance parsing pipeline for complex documents like PDFs, scans, and scientific papers. The workflow uses Docling for layout detection, Qwen3-VL-8B for describing figures/tables concurrently, and Ollama for embeddings. Built with stateless Celery workers calling GPU services, it achieves 118.7 pages/min on a single H200. The repo includes full runbooks and tuning notes.
More from coding & agent
- Claude + Thrixel build full Roblox games from a single prompt — RanaHanocka · 2026-09-02
- Vespper launches DOCX MCP for faster, cheaper agent doc editing — ycombinator · 2026-09-02
- Reef: Open-Source Infra for Self-Improving Agents via Inference Signals — pliang279 · 2026-09-02
- Agent retried for 16 hours silently? Author introspects durable runtime with Postgres persistence — slateraligator · 2026-09-02
- Copilot vs Claude Code: Which burns fewer tokens under identical conditions? — alex_bababu · 2026-09-02
- My coding agent hides dead code. This 7-second check finds it all. — timhartmann7 · 2026-09-02