ABOUT THIS MODEL

A 256M-parameter vision-language model for document understanding. Handles OCR, layout analysis, table extraction, and visual QA — all without a GPU. Built for regulated industries that process high volumes of documents privately.

KEY HIGHLIGHTS
  • Runs entirely on CPU — no GPU costs
  • Processes invoices, receipts, forms, and contracts
  • 698+ downloads on HuggingFace
  • Designed for air-gapped and regulated environments
SPECIFICATIONS
Parameters256M
ArchitectureSigLIP + SmolLM2
TasksOCR · Layout · VQA · Table Extraction
HardwareCPU-runnable · No GPU required
FormatHuggingFace Transformers · ONNX
LicenseApache 2.0
Downloads698+
QUICK START

Get started in minutes

Install
pip install transformers torch pillow
Usage
from transformers import AutoProcessor, AutoModelForVision2Seq
from PIL import Image

processor = AutoProcessor.from_pretrained("eulogik/TinyDoc-VLM-256M")
model = AutoModelForVision2Seq.from_pretrained("eulogik/TinyDoc-VLM-256M")

image = Image.open("document.png")
inputs = processor(images=image, text="Extract all text", return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=512)
print(processor.decode(output[0], skip_special_tokens=True))

Need help deploying TinyDoc-VLM?

We offer consulting, custom fine-tuning, and on-premise deployment for all our models.