---
license: apache-2.0
pipeline_tag: automatic-speech-recognition
language:
  - en
  - fr
  - de
  - es
  - it
  - pt
  - nl
  - pl
  - ru
  - ja
  - ko
  - zh
  - ar
  - hi
tags:
  - executorch
  - metal
  - asr
  - speech-to-text
  - voxtral
base_model: mistralai/Voxtral-Mini-3B-2507
---

# Voxtral Mini (3B) — ExecuTorch Export

This repository contains an ExecuTorch `.pte` export of the Mistralai Voxtral-Mini-3B-2507 model [`mistralai/Voxtral-Mini-3B-2507`](https://huggingface.co/mistralai/Voxtral-Mini-3B-2507) for:

- **Metal** (macOS): **bf16**

## Artifacts

| File | Description |
|---|---|
| `model.pte` | ExecuTorch program (main model) |
| `voxtral_preprocessor.pte` | ExecuTorch program (audio preprocessor) |
| `tekken.json` | Tokenizer configuration |

## Model Details

- **Base Model**: [`mistralai/Voxtral-Mini-3B-2507`](https://huggingface.co/mistralai/Voxtral-Mini-3B-2507)
- **Backend**: Metal (MPS)
- **Precision**: bf16
- **Parameters**: 3B
- **Languages**: Multilingual (English, French, German, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Korean, Chinese, Arabic, Hindi)

## How to Run (ExecuTorch)

Build the Voxtral runner from an ExecuTorch checkout:

```bash
EXECUTORCH_BUILD_KERNELS_TORCHAO=1 TORCHAO_BUILD_EXPERIMENTAL_MPS=1 ./install_executorch.sh
make voxtral-metal
```

Then run inference:

### Metal (macOS)

```bash
./cmake-out/examples/models/voxtral/voxtral_runner \
  --model_path /path/to/model.pte \
  --preprocessor_path /path/to/voxtral_preprocessor.pte \
  --audio_path /path/to/audio.wav \
  --tokenizer_path /path/to/tekken.json
```

## License / Attribution

- Upstream model license: **Apache-2.0** (see [`mistralai/Voxtral-Mini-3B-2507`](https://huggingface.co/mistralai/Voxtral-Mini-3B-2507)).
- Please attribute Mistral AI when using these exported artifacts.