Documents / IAM lines / full 2,915-line test
Handwriting recognition / full IAM test
Handwritten line image → text
Model size vs score
Static, watermarked figures. SlayerLab models are red. The close-up expands the crowded small-model cluster without changing any coordinates.
Measured results
Full pinned test split · 2,915 examples · cer lower is better. No global SOTA claim.
| System | Score | Samples | Elapsed¹ | Evidence |
|---|---|---|---|---|
| microsoft/trocr-large-handwritten e68501f437cd | 3.4% | 2915 | 285s | Inspect run → 18 Oct 2026, 18:08 UTC |
| microsoft/trocr-base-handwritten eaacaf452b06 | 4.7% | 2915 | 197s | Inspect run → 18 Oct 2026, 18:06 UTC |
| microsoft/trocr-small-handwritten b4648cfa1719 | 6.3% | 2915 | 44s | Inspect run → 18 Oct 2026, 18:03 UTC |
¹ Evaluation elapsed time includes model loading after dataset selection. It is not serving latency or a fair speed comparison.
What this measures
Benchmark origin
Dataset / split
Teklia/IAM-line / test
Immutable dataset revision
fbdad97500ce54635c0d1ba306bf535cb40656cfUpstream license metadata
mit · consult upstream terms
Protocol
codesota-ocr-iam-full-batch8-v1 / ocr
Selection
All 2915 examples in the pinned test split. SNLI rows without a valid label are excluded.
Configuration
Seed 42. Greedy generation for math and code. Mean answer-token log likelihood for text classification, completion and declarative multiple-choice recipes. No model judge.
Planned runs
0 qualified visits in 30 days · priority 35.1. Interest changes execution order, within the approved baseline list.
| System | Status |
|---|---|
| microsoft/trocr-small-handwritten | completed |
| microsoft/trocr-base-handwritten | completed |
| microsoft/trocr-large-handwritten | completed |