TLDR Tech / Data
How we OCR'ed 30,000 papers using Codex, open OCR models, and Jobs
Hugging Face used an open OCR model (Chandra-OCR-2) and Codex-generated scripts on serverless GPUs to convert ~27,000 papers into Markdown so they can be “chat with paper”-enabled. Running jobs in parallel made the process fast (~30 hours) and relatively cost-efficient (~$850).
Read the full story