---
license: mit
language:
- en
pipeline_tag: image-to-text
datasets:
- katanaml-org/invoices-donut-data-v1
---

## Sparrow - Data extraction from documents with ML

This model is finetuned Donut ML base model on invoices data. Model aims to verify how well Donut performs on enterprise docs.

Mean accuracy on test set: 0.96

Inference:

![Inference Results](https://raw.githubusercontent.com/katanaml/sparrow/main/sparrow-ui/assets/inference_actual.png)

Training loss:

![Training Loss](https://raw.githubusercontent.com/katanaml/sparrow/main/sparrow-ui/assets/donut_training_loss.png)

Sparrow on [GitHub](https://github.com/katanaml/sparrow)

Sample invoice [docs](https://github.com/katanaml/sparrow/tree/main/sparrow-ui/docs/images) to use for inference (docs up to 500 were used for fine-tuning, use docs from 500 for inference)

Our website [KatanaML](https://www.katanaml.io)

On [Twitter](https://twitter.com/katana_ml)