README.md

---
license: apache-2.0
base_model: utter-project/EuroMoE-2.6B-A0.6B-Instruct-2512
tags:
  - mlx
  - 4-bit
language:
  - multilingual
library_name: mlx
---

# EuroMoE-2.6B-A0.6B-Instruct-2512-MLX-4bit

MLX 4-bit quantisation of [`utter-project/EuroMoE-2.6B-A0.6B-Instruct-2512`](https://huggingface.co/utter-project/EuroMoE-2.6B-A0.6B-Instruct-2512),
converted for use on Apple Silicon via [`mlx-lm`](https://github.com/ml-explore/mlx-lm).

## Source model

- **Repository**: [`utter-project/EuroMoE-2.6B-A0.6B-Instruct-2512`](https://huggingface.co/utter-project/EuroMoE-2.6B-A0.6B-Instruct-2512)
- **Release**: 2025-12
- **Family**: eurollm
- **Origin**: eu
- **Languages / coverage**: 35 languages, same EuroLLM EU coverage. Sparse MoE: 2.6B total params, 0.6B active per token.
- **License**: apache-2.0 (inherited)


## Notes from upstream

Mixture-of-experts variant from the EuroLLM team. Same EuroBlocks instruction tuning. config.json declares model_type=mixtral with 64 experts / 8 active per token, so mlx-lm's mixtral.py handles it.


## Conversion details

- **Tool**: `mlx-lm` 0.31.3
- **Quantisation**: 4-bit (defaults from `mlx_lm.convert`)
- **Converted on**: 2026-05-05

## Usage

```python
from mlx_lm import load, generate

model, tokenizer = load("luiscalisto/EuroMoE-2.6B-A0.6B-Instruct-2512-MLX-4bit")
prompt = "Hello, who are you?"
print(generate(model, tokenizer, prompt=prompt, max_tokens=128, verbose=False))
```

## License and attribution

This is a quantised redistribution of `utter-project/EuroMoE-2.6B-A0.6B-Instruct-2512`. The original model and
its license terms (apache-2.0) carry through unchanged. Please cite the
upstream authors when using this model. See the source repository for the
authoritative model card and citation.

## Conversion provenance

Produced by [`llm-mlx-conversions`](https://codeberg.org/lcalisto/llm-mlx-conversions),
a small utility for publishing community MLX 4-bit quants of open-weight LLMs.