savitr — fast Surya OCR on Apple Silicon, for Indian electoral rolls

savitr runs Surya OCR (datalab-to/surya-ocr-2) on Apple Silicon via MLX. It also ships an electoral-roll-specific model that emits one compact line per voter, plus a pipeline that turns scanned roll PDFs into the canonical voter CSV. It runs locally without a cloud GPU.

Install

Requirements: an Apple-Silicon Mac (M-series — the OCR runs on MLX) and poppler (used to read PDFs):

brew install poppler             # macOS  (Debian: sudo apt-get install poppler-utils)

Then:

pip install savitr               # MLX runtime + terse roll model (auto-downloaded from HF)
pip install "savitr[train]"      # + the distillation toolchain (transformers/peft/torch)

Latest from git: pip install "git+https://github.com/in-rolls/savitr"

The terse model is fetched from gojiberries/savitr on first use (~800 MB, one time). The pure-Python parsing API (parse_terse) imports on any platform; only the OCR itself needs Apple Silicon.

Quickstart

The distilled terse model is the default and auto-downloads on first run — a bundled sample roll lets you try it immediately:

# OCR the bundled sample roll -> per-page voter summary (works from any install)
savitr ocr "$(savitr sample)"

# ... or write the voter records straight to CSV
savitr ocr "$(savitr sample)" -o voters.csv

# whole rolls -> canonical voter CSV (a directory of *_ENG.pdf, or a single -f file)
savitr parse-rolls -d english/ -o voters.csv

from savitr import MLXSuryaOCR, parse_terse, resolve_terse_model
from savitr.rolls.parse import TERSE_PROMPT

eng = MLXSuryaOCR(
    resolve_terse_model(), prompt=TERSE_PROMPT
)  # downloads the model if not local
text, _ = eng.ocr_image("page.png")
voters = parse_terse(text)  # [{'id': 'KMY...', 'elector_name': ..., 'age': ..., ...}]

Two models, and which one you want

Electoral rolls → the terse model. Published at gojiberries/savitr and downloaded on first use, so everything above works with no setup. It was distilled to emit voter rows and will emit them whatever the page holds — it is not a general OCR.

Anything else → base Surya. Upstream publishes datalab-to/surya-ocr-2 but not an MLX build of it, so convert it once (~1.3 GB fetched, ~500 MB written):

python -m mlx_vlm convert --hf-path datalab-to/surya-ocr-2 \
    --mlx-path models/surya-mlx-4bit -q --q-bits 4

from savitr import MLXSuryaOCR

eng = MLXSuryaOCR()  # finds models/surya-mlx-4bit, or $SAVITR_BASE_PATH
eng = MLXSuryaOCR("/some/other/model")  # or say where
text, _ = eng.ocr_image("page.png")  # HTML: <table><tr><td>…

With no converted model, every entry point — the constructor, savitr ocr --html, and --cover-model — says the same thing and repeats the command above. savitr ocr --html needs it; --cover-model is optional metadata and carries on without it.

terse-Surya (gojiberries/savitr)

Surya self-distilled to emit pipe-delimited voter rows. The model card is the single source for checkpoint provenance, evaluation definitions, results, and limitations.

What's in the box

Three layers — two you install and use, one for reproducing the model:

src/savitr/            # the pip package (use it)
  mlx_ocr.py           # GENERIC MLX Surya engine (MLXSuryaOCR) — run any Surya OCR fast
  rolls/               # ELECTORAL-ROLL app: parse · fields · schema · pipeline · ocr
  cli.py, __init__.py
training/              # repo-only: build_corpus · train · eval · merge · kaggle_*  (reproduce the model)

• Use it — electoral rolls (the product): savitr parse-rolls / savitr ocr run the distilled terse model (the default) on roll PDFs → voter records / canonical CSV (savitr.rolls).
• Use it — generic fast Surya: savitr.MLXSuryaOCR runs any compatible Surya model on Apple Silicon.
• Reproduce it — training/distillation: lives in top-level training/, not shipped in the wheel (install the [train] extra to run it). We ship code to use the model, not to train it.

How it was built / what was tried

See the findings for the measured baseline (decode, not cold-start, dominates; ~110 s/page on llama.cpp), the MLX win, the tested negatives (compact prompt, guided JSON, parallelism), and the distillation method and numbers.

Develop

uv sync --all-groups --all-extras
uv run pytest
uv run ruff check .
uv run ruff format --check .