📄 TinyDoc — AI Document Understanding in 3 Lines of Code

Extract answers, tables, and structured data from any document image.

What is this?

TinyDoc is a Python SDK powered by TinyDoc-VLM — a 256M parameter vision-language model trained specifically for document understanding. It runs on CPU with no GPU required.

Drop in a document image. Ask a question. Get the answer.

Install

pip install tinydoc

That's it. One package. ~1.1GB model weights auto-download from HuggingFace on first use.

Quick Start

from PIL import Image
from tinydoc import TinyDocExtractor

extractor = TinyDocExtractor()  # auto-detects device, loads from HF Hub
img = Image.open("invoice.png")

# 💬 Ask a question
result = extractor.ask(img, "What is the total amount?")
print(result.answer)  # "$1,234.56"

# 📋 Extract all fields as JSON
result = extractor.extract(img, output_format="json")
print(result.fields)  # {"vendor": "Acme Corp", "total": "$1,234.56", "date": "2024-01-15", ...}

# 📊 Extract tables to Markdown
result = extractor.extract_table(img)
print(result.markdown)
# | Item       | Qty | Price  |
# |------------|-----|--------|
# | Widget A   | 10  | $25.00 |
# | Widget B   | 5   | $50.00 |

What can it do?

Why TinyDoc?

Advanced

extractor = TinyDocExtractor(
    device="cuda",           # or "cpu", "mps"
    model_name_or_id="eulogik/TinyDoc-VLM-256M",  # or local path
)

result = extractor.ask(
    img,
    "What are the line items?",
    max_new_tokens=256,      # override default 512
)

Links

License

Apache 2.0 — free for commercial use.

Built by eulogik — AI infrastructure for document intelligence.