visiontest_a_s1338_d20m
d20m (27.1M parameters) on plus_clean subset, trained on RTX 3060 12GB (Home GPU). Started 3 Oct 2026, ended 3 Oct 2026.
Hypothesis
Control: a clean-text subset with no OCR, at 20M.
What we learned
The reference for arms B to D.
Scores
Bits per byte, lower is better; change against the parent run, VISION-OCR A.
ex-Gītā (headline)
0.7937
bits per byte
+0.3%ex-Gītā, clean_v1
—
bits per byte
Pooled, all five sets
0.7878
bits per byte
+0.3%Validation split
0.7828
bits per byte
−0.1%| Test set | VISION-OCR A s2 | VISION-OCR A (parent) | Change |
|---|---|---|---|
| DCS gold (classical) | 0.7769 | 0.7758 | +0.1% |
| Bhagavad-gītā (memorisation) | 0.6334 | 0.6338 | −0.1% |
| Out of domain | 0.8001 | 0.7976 | +0.3% |
| Prose | 0.7578 | 0.756 | +0.2% |
| Vedic (Ṛgveda) | 1.1005 | 1.0937 | +0.6% |
Curves
No training curves were published for this run.Its scores above are what it published.
Model
- Preset
- d20m
- Parameters
- 27,073,024
- Outside embeddings
- 18,881,024
- Layers · heads · width
- 6 · 8 · 512
- Tokenizer
- SLP1 unigram 8k
Data
- Slice
- plus_clean subset
- Words
- —
- Training tokens
- 89,885,258
- Passes
- 1 passes
- Tokens seen
- 89,899,008
Compute
- GPU
- RTX 3060 12GB
- Where
- Home GPU
- Steps
- 1,829 / 1,829
- GPU hours
- 0.23
- Cost
- —
- Spot restarts
- —
Lineage
homed20m