File size: 3,580 Bytes
bd063c2 5321d8a bd063c2 02c76cd bd063c2 5321d8a bd063c2 d22dd62 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 | ---
base_model: mistralai/Voxtral-Mini-4B-Realtime-2602
library_name: mlx
license: apache-2.0
pipeline_tag: automatic-speech-recognition
tags:
- voxtral
- audio
- speech
- speech-recognition
- realtime
- streaming
- asr
- mlx
- turboquant
- quantization
- 4-bit
---
> [!TIP]
> **KV-cache quantization without any fork (recommended, 2026):** upstream
> llama.cpp/Ollama now cover this natively — use `-ctk q8_0 -ctv q8_0`
> (~half KV memory, negligible quality loss: perplexity +0.002–0.05) or
> `-ctk q4_0 -ctv q4_0` (~quarter memory, ≈7.6% perplexity increase). In
> Ollama: `OLLAMA_KV_CACHE_TYPE=q8_0` with `OLLAMA_FLASH_ATTENTION=1`. Keep
> K and V types symmetric to stay on the fast fused Flash-Attention path.
> Since April 2026, mainline llama.cpp also applies Hadamard rotation to
> KV activations ([PR #21038](https://github.com/ggml-org/llama.cpp/pull/21038)),
> which greatly improves low-bit KV quality (opt-out:
> `LLAMA_ATTN_ROT_DISABLE=1`).
>
> The RotorQuant/TurboQuant fork flow below is **experimental/legacy**: the
> TurboQuant llama.cpp PR was closed without merging (June 2026) and the fork
> is unmaintained relative to mainline. It is NOT required to use this model.
<!-- kv-upstream-note -->
# Voxtral-Mini-4B-Realtime-2602-TurboQuant-MLX-4bit
4-bit MLX weight-quantized build of [`mistralai/Voxtral-Mini-4B-Realtime-2602`](https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602) with TurboQuant KV-cache. Recommended default for real-time ASR on Apple Silicon.
## Hardware compatibility
| Device | VRAM / RAM | Recommendation |
| --- | --- | --- |
| Apple M4 Max 128 GB | ~2.6 GB | recommended — headroom for long context |
| Apple M3 Max 64 GB | ~2.6 GB | comfortable |
| Apple M2 Max 32 GB | ~2.4 GB | fits |
## Overview
- **Base:** `mistralai/Voxtral-Mini-4B-Realtime-2602` — 4B real-time ASR model
- **Weight precision:** 4-bit (group-wise)
- **KV-cache profile:** TurboQuant
- **Approx. on-disk size:** ~2 GB
- **Runtime:** MLX on Apple Silicon
## Quickstart
```bash
pip install mlx-lm
```
```python
from mlx_lm import load, generate
model, tokenizer = load("majentik/Voxtral-Mini-4B-Realtime-2602-TurboQuant-MLX-4bit")
for chunk in audio_stream():
prompt = tokenizer.apply_chat_template(
[{"role": "user", "content": [{"type": "audio", "path": chunk}]}],
add_generation_prompt=True,
)
emit(generate(model, tokenizer, prompt=prompt, max_tokens=32))
```
## Model specs
| Field | Value |
|---|---|
| Parameters | 4B |
| Weight bits | 4 |
| Group size | 64 |
| Cache profile | TurboQuant |
| Size on disk | ~2 GB |
| Target hardware | Apple Silicon (M1/M2/M3/M4) |
| License | Apache 2.0 |
## RotorQuant vs TurboQuant
| | TurboQuant | RotorQuant |
|---|---|---|
| Strategy | Per-head static calibration | Rotational online re-basis |
| Memory reduction | ~3.5x on KV-cache | ~4x on KV-cache |
| Best for | Predictable domains, lowest p50 latency | Noisy/multi-speaker streams |
## See also
- [`majentik/Voxtral-Mini-4B-Realtime-2602-TurboQuant-MLX-8bit`](https://huggingface.co/majentik/Voxtral-Mini-4B-Realtime-2602-TurboQuant-MLX-8bit)
- [`majentik/Voxtral-Mini-4B-Realtime-2602-TurboQuant-MLX-2bit`](https://huggingface.co/majentik/Voxtral-Mini-4B-Realtime-2602-TurboQuant-MLX-2bit)
- [`majentik/Voxtral-Mini-4B-Realtime-2602-RotorQuant-MLX-4bit`](https://huggingface.co/majentik/Voxtral-Mini-4B-Realtime-2602-RotorQuant-MLX-4bit)
- [`mistralai/Voxtral-Mini-4B-Realtime-2602`](https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602) — upstream base model |