Re-OCR of 14,483 pages on one GPU for $1.47: Saturate batch inference hits 8.9 pages/s
vanstriendaniel · x · 2026-11-02
Developer vanstriendaniel re-OCR'd 14,483 pages of fossil fuel industry documents with his library Saturate: one GPU, $1.47 total, 8.9 pages/s once running via Hugging Face Jobs — concurrency self-settled at 512 with no tuning.
Saturate is batch inference for datasets: point it at any OpenAI-compatible endpoint (your own vLLM server, SGLang Job, TEI, an inference endpoint) and it runs every row or bucket file through the model, outputting resumable parquet rather than a pile of responses, with congestion-aware concurrency. The attached image shows existing OCR mangling a page while Xiaomi-OCR-0 reads it nearly correctly. Open-sourced on GitHub under Apache-2.0.
Related event: Single GPU OCRs 14,483 Pages for Just $1.47(2 posts)→
More from coding & agent
- Dev's AI comedy generator produces decent jokes 15-20% of the time — ivan_bezdomny · 2026-11-03
- Dev builds SwitchVoice in Cursor one-shot to dub YouTube videos in cloned voices — Nilofer_tweets · 2026-11-03
- Meta paper: routing agent memory by task type beats shared history, cuts API cost 65-72% — rohanpaul_ai · 2026-11-03
- Prototype Turns Claude Code Answers Into 1939-Style Cartoon Explainer Videos via H3 Max — OdinLovis · 2026-11-03
- Meituan LongCat's Three-Font Bug: A font Stack Built Only for Chinese Windows — teortaxesTex · 2026-11-03
- Career hack: work where leadership funds Claude but doesn't grasp its potential — brandon_galang · 2026-11-03