README.md

---
title: README
emoji: 🎵
colorFrom: red
colorTo: pink
sdk: static
pinned: false
---

<h1 align="center">Musixmatch</h1>

<p align="center">
  Musixmatch is the World’s Leading Music Data Company. <br/ >
  Our mission is to provide data, tools and services that allows the experience of music to be enriched across the whole world.
</p>

<p align="center">
  <a href="https://www.musixmatch.com">Website</a> ·
  <a href="https://about.musixmatch.com">About</a> ·
  <a href="https://developer.musixmatch.com">Developer API</a> ·
  <a href="https://github.com/musixmatchresearch">GitHub</a> ·
  <a href="https://about.musixmatch.com/careers">Careers</a>
</p>

---

## AI research

Musixmatch best-known open contribution is **UmBERTo**, a RoBERTa-based Italian language
model trained with SentencePiece tokenization and Whole Word Masking on 70GB of
deduplicated Italian text from the OSCAR corpus (~11B words). It set a new bar
for Italian NLP and remains a widely used baseline.

| Model | Task | Score |
|---|---|---|
| UmBERTo (CommonCrawl, cased) | NER — WikiNER-ITA | **92.5 F1** |
| UmBERTo (CommonCrawl, cased) | NER — ICAB-EvalITA07 | **87.6 F1** |
| UmBERTo (CommonCrawl, cased) | POS — UD Italian-ISDT | **98.9 acc** |

→ [`Musixmatch/umberto-commoncrawl-cased-v1`](https://huggingface.co/Musixmatch/umberto-commoncrawl-cased-v1)
· [`Musixmatch/umberto-wikipedia-uncased-v1`](https://huggingface.co/Musixmatch/umberto-wikipedia-uncased-v1)
· [source on GitHub](https://github.com/musixmatchresearch/umberto)

We also run our training and inference on AWS — read the
[Musixmatch × Hugging Face × AWS case study](https://huggingface.co/case-studies/aws/musixmatch).

<p align="center">
  <sub>© Musixmatch · Bologna, Italy</sub>
</p>