skip to content
gigarouter gigarouter
models / image-to-text · coming soon

Donut Base

naver-clova-ix/donut-base

published Jul 2022 · updated Aug 2022

Donut Base is an image-to-text model that performs OCR-free document understanding using a Swin Transformer encoder and a BART decoder.

status
coming soon
API providers
0
downloads / mo
166K
license
mit

specs

TaskImage-to-Text / Document Understanding
ArchitectureSwin Transformer encoder + BART decoder

about this model

Donut (base-sized) is an image-to-text model that performs OCR-free document understanding using a vision encoder (Swin Transformer) and a text decoder (BART). Given an image, the encoder produces a sequence of embeddings, and the decoder autoregressively generates text conditioned on those embeddings.

Architecture diagram of Donut model showing Swin Transformer encoder and BART decoder

Introduced in the paper OCR-free Document Understanding Transformer (ECCV 2022), Donut eliminates the need for external OCR engines, reducing computational cost, increasing flexibility across languages and document types, and avoiding OCR error propagation. According to the paper, Donut achieves state-of-the-art results on multiple visual document understanding (VDU) benchmarks in both speed and accuracy.

This pre-trained base model is designed to be fine-tuned on downstream tasks such as document image classification or information extraction (document parsing). When hosted via gigarouter’s API, it provides a foundation for building custom document understanding pipelines without managing infrastructure.

best for

FAQ

What is the main advantage of Donut over OCR-based approaches?

Donut eliminates the need for external OCR engines, reducing computational cost and avoiding error propagation.

Is this model ready for inference, or does it require fine-tuning?

This is a pre-trained only model; it must be fine-tuned on a downstream task such as document classification or parsing.

What are the input and output formats for this model?

Input: an image (e.g., a document scan). Output: generated text (e.g., structured JSON for parsing, or a label for classification).

How can I use this model via the gigarouter API?

Use the OpenAI-compatible endpoint with your API key; send the image as a base64-encoded string in the request.

not yet live

We're benchmarking and onboarding Donut Base as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.

related image-to-text models

compare all →