0.8B Qwen Fine-Tuned on Medieval Manuscripts Sparks CER Debate

A researcher fine-tuned a 0.8B Qwen 3.5 on over 2,000 medieval manuscripts totaling 10,000 pages. The project sparked debate over whether CER remains a valid metric for full-page OCR, as layout and reading-order decisions get counted as reading errors.

2026-09-03 ~ 2026-09-04 · 4 related posts

• CER no longer fits full-page OCR: layout and style decisions all count as errors — staghado · 2026-09-03
• 0.8B Qwen 3.5 full finetune transcribes 2,000+ medieval manuscripts across 10,000 pages — wjb_mattingly · 2026-09-04
• Why CER still matters for manuscript transcription models, despite its limits — wjb_mattingly · 2026-09-04

1 near-duplicate retellings: vanstriendaniel