chandra

Features

• Convert documents to markdown, html, or json with detailed layout information
• Good handwriting support
• Reconstructs forms accurately, including checkboxes
• Good support for tables, math, and complex layouts
• Extracts images and diagrams, with captions and structured data
• Support for 40+ languages

Quickstart

The easiest way to start is with the CLI tools:

pip install chandra-ocr

# With VLLM
chandra_vllm
chandra input.pdf ./output

# With HuggingFace
chandra input.pdf ./output --method hf

# Interactive streamlit app
chandra_app

Benchmarks

We used the olmocr benchmark, which seems to be the most reliable current OCR benchmark in our testing.

Examples

Usage

Installation

pip install chandra-ocr

From code

from chandra.model import InferenceManager
from chandra.model.schema import BatchInputItem

# Run chandra_vllm to start a vLLM server first if you pass vllm, else pass hf
# you can also start your own vllm server with the datalab-to/chandra model
manager = InferenceManager(method="vllm")
batch = [
    BatchInputItem(
        image=PIL_IMAGE,
        prompt_type="ocr_layout"
    )
]
result = manager.generate(batch)[0]
print(result.markdown)

With transformers

from transformers import AutoModel, AutoProcessor
from chandra.model.hf import generate_hf
from chandra.model.schema import BatchInputItem
from chandra.output import parse_markdown

model = AutoModel.from_pretrained("datalab-to/chandra").cuda()
model.processor = AutoProcessor.from_pretrained("datalab-to/chandra")

batch = [
    BatchInputItem(
        image=PIL_IMAGE,
        prompt_type="ocr_layout"
    )
]

result = generate_hf(batch, model)[0]
markdown = parse_markdown(result.raw)

Credits

Thank you to the following open source projects:

• Huggingface Transformers
• VLLM
• olmocr
• Qwen 3 VL