CatalogWantedMirrorsLedgerOperating StatementDocsStudio statusCard system
alt.freeweights.mistral / Voxtral-Mini-4B-Realtime-2602-4bit
From: mlx-community <huggingface.co/mlx-community>Newsgroups: alt.freeweights.mistralSubject:

Voxtral-Mini-4B-Realtime-2602-4bit

Date: 2026-11-15Commit: fdebf7bSigned: 2DC28CC7B26A283B ✓

Automatic speech recognition model with 4.4B parameters. Safetensors weights, 3.15 GB in all. Published by mlx-community under Apache-2.0.

4.4B params3.15GBVoxtral-Mini-4B-Realtime-2602-4bitsafetensorspermissive
FactsParams4.4BSize3.15 GBFormatssafetensorsTaskautomatic speech recognitionLicenseApache-2.0From sourcepermissive
$ fw get mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit
# manifest sha256 89cf4ad7688a… · 3.15 GB · 6 files
$ fw verify ./Voxtral-Mini-4B-Realtime-2602-4bit mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit
From sourceHosted (signed)mlx-community/Voxtral-Mini-4B-Realtime-2602-4bitLast check: 2026-11-15 04:59 UTC
Run at homeGet help

Get help: custom fine-tune or private deploy. We review your setup before we quote.

Quantized of Voxtral-Mini-4B-Realtime-2602.

Works with

Run23Fine-tune7Convert8Scan2
OverviewModel cardFilesUse itLicenseProvenanceModel tree

Overview

Size
3.15 GB
Weights
3.13 GB (safetensors)
Parameters
4.4B
Formats
safetensors
Task
automatic speech recognition
License
Apache-2.0 (permissive)
Files
6, each with a sha256
Publisher
mlx-community
Added
2026-11-15
Section
alt.freeweights.mistral
Lineage

Original release from the publisher.

Base model: mistralai/Voxtral-Mini-4B-Realtime-2602

Where these bytes come from
Publisher at the pinned commit→You

Sources are tried in order. A file counts only when its size and sha256 match the manifest. Sources are checked when you ask for them, not on a timer.

Sources
BadgeHostLast checkResult
From sourcehuggingface.co2026-11-15 04:59 UTCpassed

Model card

From the publisher at commit fdebf7b. Shown as plain text with links; images open at the publisher.

Voxtral Mini 4B Realtime#

This is a 4-bit quantized MLX conversion of mistralai/Voxtral-Mini-4B-Realtime-2602, Mistral AI's streaming speech-to-text model.

Runs via mlx-audio.

Key Details#

Parameters4B (~3.4B LM + ~0.6B Audio Encoder)
Quantizationint4
Base modelmistralai/Voxtral-Mini-4B-Realtime-2602
Languages13 (Arabic, German, English, Spanish, French, Hindi, Italian, Dutch, Portuguese, Chinese, Japanese, Korean, Russian)
LicenseApache 2.0

See also: fp16 variant

Usage#

pip install mlx-audio[stt]
from mlx_audio.stt.utils import load

model = load("mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit")

# Transcribe audio
result = model.generate("audio.wav")
print(result.text)

# Streaming transcription
for chunk in model.generate("audio.wav", stream=True):
    print(chunk, end="", flush=True)

# Adjust transcription delay (lower = faster but less accurate)
result = model.generate("audio.wav", transcription_delay_ms=480)
SettingValueNotes
Temperature0.0Always use greedy decoding
Transcription delay480msSweet spot of accuracy vs. latency
Delay range80ms – 2400msMultiples of 80ms

Benchmarks (from upstream)#

FLEURS (13 languages, WER%)#
DelayAVGENFRDEESZHJAKO
160ms12.606.469.759.505.3417.6719.1719.81
480ms8.724.906.426.193.3110.459.5915.74
960ms7.704.345.684.872.988.996.8014.90
2400ms6.734.055.234.152.718.485.5014.30
Long-form English (WER%)#
DelayMeanwhileEarnings-21Earnings-22TEDLIUM
480ms5.0510.2312.303.17

Architecture#

  • Causal audio encoder (~0.6B) with sliding window attention — enables true streaming
  • Language model decoder (~3.4B) based on Ministral-3B with adaptive RMS norm conditioned on transcription delay
  • 4x downsampling from encoder to decoder (frame rate = 12.5 Hz)
  • Both components use sliding window attention for unbounded audio length

More Info#

Files

6 files · 3.15 GBWeights 3.13 GB (safetensors). Sizes and sha256 from the signed manifest.
Hide all 6 files with their sha256
FileSizesha256
.gitattributes1.57 KBa1533fad1fb49e1230aff61a60028ea63bc924b4cc4501447351b50b9c840683
README.md3.13 KB022af0f470bbb2036d1d8fea62118d1d14f64b9cbb38415004544990e653f99e
config.json1.51 KB02060864a4f33df5e4944684fc17f3026af4011830cac4def6e9e025315b10c5
model.safetensors3.13 GB6f59b425d8a1ceb2de795454558be63937cf75b59f9c9bc77accd85aaf32af05
model.safetensors.index.json119 KB80f68b80cf4b1638d864d1504061a266f59e37a8d90d7b20f2e1f30c2d034c2e
tekken.json14.9 MB8434af1d39eba99f0ef46cf1450bf1a63fa941a26933a1ef5dbbf4adf0d00e44
VersionsEach time the pinned upstream commit or file list changed in this repo. Newest first.
Upstream commitDateContents
fdebf7b2af pinned now2026-11-15 04:59 UTC6 files · 3.15 GB

Use it

$ fw get mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit
# manifest sha256 89cf4ad7688a… · 3.15 GB · 6 files
$ fw verify ./Voxtral-Mini-4B-Realtime-2602-4bit mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit
fw.pyChecks the signature and every sha256, tries each source in order, resumes.
$ python3 fw.py get https://freeweights.allternit.com/manifests/mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit.json --dir ./Voxtral-Mini-4B-Realtime-2602-4bit $ python3 fw.py verify ./Voxtral-Mini-4B-Realtime-2602-4bit ./mlx-community__Voxtral-Mini-4B-Realtime-2602-4bit.json
hfThe Hugging Face CLI with one setting. Same repo id.
$ HF_ENDPOINT=https://freeweights.allternit.com hf download mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit --revision fdebf7b2af834a1db4b8a3c99ab7480b333adf9e --local-dir ./Voxtral-Mini-4B-Realtime-2602-4bit
transformersSpeech recognition with the transformers pipeline.
import os os.environ["HF_ENDPOINT"] = "https://freeweights.allternit.com" from transformers import pipeline asr = pipeline("automatic-speech-recognition", model="mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit", revision="fdebf7b2af834a1db4b8a3c99ab7480b333adf9e") print(asr("audio.wav")["text"])
Run at home

License

Name
Apache License 2.0
Id
Apache-2.0
Class
Permissive
Redistribution
yes
Listing
Hosted (signed)
Decided
License as declared by the publisher. No person read it. 2026-11-15
Text in
README.md
Text sha256
022af0f470bbb2036d1d8fea62118d1d14f64b9cbb38415004544990e653f99e
Terms gate
No extra terms beyond the license.

Read the license text

Licenses are checked against the license table. A custom license passes only after a person reads it.

Provenance

From: mlx-community <huggingface.co/mlx-community>Newsgroups: alt.freeweights.mistralSubject: Voxtral-Mini-4B-Realtime-2602-4bitDate: 2026-11-15Commit: fdebf7bSigned: 2DC28CC7B26A283B ✓
Upstream
mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit
Pinned commit
fdebf7b2af834a1db4b8a3c99ab7480b333adf9e
Manifest
Voxtral-Mini-4B-Realtime-2602-4bit.json · .minisig
Manifest sha256
89cf4ad7688a6fa1ca98fca27e173c5a78996f0c0308013b177ced96ad196e65
Signature
minisign, Ed25519, key 2DC28CC7B26A283B
Created
2026-11-15 04:59 UTC
Generator
freeweights-manifest/0.1.0
Takedowns
None.

How to verify

$ curl -O https://freeweights.allternit.com/manifests/mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit.json $ curl -O https://freeweights.allternit.com/manifests/mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit.json.minisig $ minisign -Vm Voxtral-Mini-4B-Realtime-2602-4bit.json -P RWQ7KGqyx4zCLXtUFOqvfv8L3g+8fpKJ1FsuxPyOYMF+4KufbUxKq1Oq $ shasum -a 256 <file> # compare with the Files tab

A manifest whose signature fails must be rejected. A file whose size or sha256 does not match must be discarded. Verify a download

Model tree

Parent

mistralai/Voxtral-Mini-4B-Realtime-2602 · this model is a quantization of it

No model in the catalog names this one as its parent.

Get it
Download
$ HF_ENDPOINT=https://freeweights.allternit.com hf download mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit
Open in Freeweights Studio

Pull, verify and run it in the Allternit app on Desktop, web or phone.

Signed manifest
minisign, key 2DC28CC7B26A283B. Voxtral-Mini-4B-Realtime-2602-4bit.json. Provenance. Takedown history: none.
Source
huggingface.co/mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit

Related models

All of alt.freeweights.mistral →
From: bartowski <huggingface.co/bartowski>Subject: alamios_Mistral-Small-3.1-DRAFT-0.5B-GGUF
From: microsoft <huggingface.co/microsoft>Subject: chatbench-mistral-7b
From: NousResearch <huggingface.co/NousResearch>Subject: DeepHermes-3-Mistral-24B-Preview
From: mistralai <huggingface.co/mistralai>Subject: Devstral-Small-2-24B-Instruct-2512
From: mlx-community <huggingface.co/mlx-community>Subject: Devstral-Small-2-24B-Instruct-2512-4bit
From: mlx-community <huggingface.co/mlx-community>Subject: Devstral-Small-2-24B-Instruct-2512-6bit