Downloads · 30 days
42
21% of all-time downloads
aufklarer/Voxtral-Mini-3B-2507-MLX-8bit
Voxtral-Mini-3B-2507-MLX-8bit is a automatic speech recognition model from aufklarer. Use it when you need speech turned into text. It is set up for mlx. The card lists the license as apache-2.0.
MLX weights for Voxtral Mini 3B 2507, ready for local speech-to-text with speech-swift.
Downloads · 30 days
42
21% of all-time downloads
All-time downloads
204
Public
Parameters
4.7B
5.6 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors5.6 GB · 100%
How the weights are stored.
U324B · 86%
From the Hugging Face model README
MLX weights for Voxtral Mini 3B 2507, ready for local speech-to-text with speech-swift.
The FP16, INT5, and INT8 family passed the isolated WER, RTF, throughput, RSS, and physical-footprint compatibility gate on 2026-07-22 using English FLEURS.
| Parameters | Precision | Sample rate | Languages |
|---|---|---|---|
| 3B | 8-bit affine | 16 kHz | 8 |
| File | Description |
|---|---|
model*.safetensors | MLX model weights |
config.json and params.json | Model and runtime configuration |
tekken.json | Tekken tokenizer data |
speech transcribe recording.wav --engine voxtral --model int8
import VoxtralASR
let model = try await VoxtralModel.load(
"aufklarer/Voxtral-Mini-3B-2507-MLX-8bit"
)
let text = model.transcribe(
audio: samples,
sampleRate: sourceSampleRate,
language: "en"
)
English FLEURS read speech, 80 utterances (759.56 seconds), Apple M5 Pro:
| Variant | Bundle | WER | Mean RTF | Peak RSS |
|---|---|---|---|---|
| FP16 | 8.71 GiB | 4.633% | 0.1305 | 9,091 MiB |
| INT5 | 3.77 GiB | 4.744% | 0.0739 | 4,035 MiB |
| INT8 | 5.18 GiB | 4.578% | 0.0906 | 5,487 MiB |
These read-speech measurements are not Artificial Analysis Conversational or ElevenLabs Scribe results.
The base model is mistralai/Voxtral-Mini-3B-2507 and is released under Apache 2.0.