savitr — terse electoral-roll OCR (distilled Surya)

gojiberries/savitr is datalab-to/surya-ocr-2 (650M Qwen3.5-VL-style OCR) self-distilled to read Indian electoral-roll pages and emit one compact, pipe-delimited line per voter instead of verbose HTML:

epic|name|relation(F/H/M)|relative_name|house|age|sex

That is ~5× fewer decode tokens than the HTML output, so it runs ~2.7× faster end-to-end at the teacher's accuracy. Converted to MLX 8-bit for Apple Silicon.

Usage

pip install savitr
savitr ocr roll.pdf --terse          # auto-downloads this model

from huggingface_hub import snapshot_download
from savitr import MLXSuryaOCR, parse_terse
from savitr.rolls.parse import TERSE_PROMPT
path = snapshot_download("gojiberries/savitr")
eng = MLXSuryaOCR(path, prompt=TERSE_PROMPT)
voters = parse_terse(eng.ocr_image("page.png")[0])

How it was trained

Teacher = full Surya (surya-ocr-2) OCRs roll pages to HTML; a parser cleans them into terse targets; the model is LoRA-fine-tuned on (page image → terse rows) — 450 pages drawn from constituencies held out of the eval, 1 epoch, for $0 on a free Kaggle T4. The terse format is the only behavioral change — reading ability is inherited from Surya.

Evaluation (out-of-sample, vs the Surya teacher)

Held-out constituencies never seen in training (37 pages, 1,076 teacher voters):

Field	Fidelity		Field	Fidelity
voter recall	99.3%		relative name	96.2%
EPIC	97.2%		relation code (F/H/M)	97.9%
name	96.2%		house	98.8%
age	97.5%		sex	98.2%

Per-voter record similarity 98.7%, whole-page similarity 92.9% (1 − normalized edit distance). Fidelity = agreement with the teacher's output; absolute accuracy ≈ these × Surya's own ~93–95%.

Limitations

v0.2 (450 training pages, AC-holdout). All fields are teacher-grade out-of-sample — EPIC, house, age, sex, the relation code, names, and recall all 96–99% (the relation code, the weak spot of the earlier 77-page model, is now 98%). Trained on Manipur 2025 English rolls; other states/scripts are out of distribution. Pair with savitr's value-anchored parse_terse, which stays column-aligned even when the model drops a field.

License & attribution

Derived from datalab-to/surya-ocr-2; its license governs use of these weights. savitr's code is MIT. Electoral rolls are public records published by the Election Commission of India.