Qwen3-VL-8B Hebrew — v1.6 checkpoints (full-page Talmud)
Cairo Genizah AI Transcription — formal analysis and write-up of the project
The full account of the methods, training data, benchmarks and error analysis behind the Cairo Genizah AI transcription models. Search the transcribed corpus at Cairo Genizah AI.
Unless you are studying this series' training history, use the flagship instead:qwen3-vl-8b-hebrew-v19a-ckpt.
Status: superseded for manuscripts; still the series' Talmud-print
reference. The full-page generation: page-level training at the 6.5 MP
resolution policy the series has used ever since.
- Official v1.6: step 1000,
d6bb5dd3c52c8048618c0b80ffece36b2556d1ed.
Talmud page-level strict CER means (65-page benchmark): gemara 0.090 /
rashi 0.047 / tosafot 0.099 — the Rashi-script column 6.7× better than
the strongest closed model we evaluated (0.315).
- Step 1100,
b9f47f32e9ed51eb45598e27ffb5c5b650f7723c (run end): the
v1.7 warm start.
- Medians across checkpoints were ceiling-level (≈0.004–0.013 per section);
checkpoint ranking is driven by collapse-page counts — see the flagship
card's notes on failure-mode-aware evaluation.
- Limitation: language-only LoRA (vision tower did not train; fixed in v1.8b).
The series at a glance
One line per generation — which checkpoint to use and which are historical:
| version |
repo |
status |
canonical revision |
headline (benchmark) |
| v1.9a |
v19a-ckpt |
FLAGSHIP — use this |
step 1300 43e21bd7 |
F1 0.816 / CER 0.216 (Genizah religious-140); F1 0.862 / CER 0.196 (frozen PGP-131) |
| v1.9b |
v19b-ckpt |
experimental control — not adopted |
step 1300 f8618c28 |
merger-LoRA ablation study; loss ≡ v1.9a |
| v1.8b |
v18b-ckpt |
superseded; warm-start ancestor of all v1.9 |
step 700 c80313f8 |
first arm with a live vision-tower LoRA |
| v1.8a |
v18a-ckpt |
superseded A/B control (vision frozen) |
step 700 |
control arm for the v1.8 vision experiment |
| v1.7 |
v17-ckpt |
superseded |
step 800 |
first Genizah-handwriting generation; best VLM on both corpora at its era's benchmarks (Aug 2026) |
| v1.6 |
v16-ckpt |
superseded; Talmud-print reference |
step 1000 (official); step 1100 b9f47f32 (v1.7 warm start) |
Talmud page CER: gemara 0.090 / rashi 0.047 / tosafot 0.099 — Rashi-script 6.7× better than the best closed model we tested (0.315) |
| v1.5 |
rashi-ckpt |
superseded |
final e050aca8 (step 2000) |
proved pure Rashi-glyph perception: CER 0.018 on unmemorizable synthetic text |
| v1 |
hebrew-ckpt |
superseded (language-only LoRA) |
step 3800 e2f85dde |
first checkpoints to read Vilna gemara from pixels |
* CER over substantive attempts only, alignment-based scorer. Benchmarks
differ across generations (Talmud print vs Genizah manuscripts) — compare
within a row's named benchmark, not across rows.
