| --- |
| title: README |
| emoji: ⚖️ |
| colorFrom: blue |
| colorTo: yellow |
| sdk: static |
| pinned: false |
| --- |
| |
| # EuLLM — sovereign LLMs for Europe |
|
|
| **Modelli linguistici aperti, verticali e locali, costruiti in Europa.** |
|
|
| EuLLM builds open-weight language models specialised for one domain and one language, small enough to run on your own hardware, and the engine to run them. No data leaves your machine. |
|
|
| ## Models |
|
|
| | Model | What it does | Size (Q4_K_M) | Held-out exam | |
| |---|---|---|---| |
| | [legal-it-8b](https://huggingface.co/EuLLM/legal-it-8b) | Answers questions on Italian law from the statutes retrieved for it, citing them | 5.2 GB | 89.0% (base model: 80.9%) | |
| | [legal-it-4b](https://huggingface.co/EuLLM/legal-it-4b) | The same, for smaller machines | 2.5 GB | 85.1% (base model: 77.6%) | |
|
|
| - **How they are used.** Both are built for retrieval-augmented, open-book use. The recommended retrieval (BM25, Qwen3-Embedding-0.6B and Qwen3-Reranker-0.6B, about 1.2 GB in all) is documented in the model cards. |
| - **How they were measured.** On 429 held-out questions that were never seen in training, with paired significance tests and with a check that needs no judge model. |
| - **Not legal advice:** they are tools for consulting the legal texts. |
|
|
| ## Engine |
|
|
| [EuLLM Engine](https://github.com/eullm/eullm) is a single Rust binary for running models like these locally: |
| - the same API as Ollama and OpenAI, so existing clients work unchanged; |
| - continuous batching; |
| - no telemetry; |
| - an audit trail designed for the EU AI Act. |
|
|
| ## Built with European compute |
|
|
| The models are trained on the Leonardo supercomputer (CINECA, Italy) through the EuroHPC AI Factory allocation EHPC-AIF-2026PG01-1147. We thank the EuroHPC Joint Undertaking for access. |
|
|
| [eullm.eu](https://www.eullm.eu) · [GitHub](https://github.com/eullm) |
|
|
|
|
