CoolFace
Modelpublic

aufklarer/Qwen3-ASR-1.7B-MLX-4bit

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face
2likes635downloads
Model Card

Qwen3-ASR-1.7B — MLX 4-bit

MLX 4-bit quantized conversion of Qwen/Qwen3-ASR-1.7B for Apple Silicon inference.

Model Details

DetailValue
ArchitectureWhisper-style audio encoder + Qwen3 text decoder
Parameters1.7B
Quantization4-bit (group_size=64, text decoder only)
Audio encoderfloat16 (24 layers, 1024 dim, 16 heads)
Size~2.1 GB
LanguagesMultilingual (EN, ZH, JA, KO, FR, DE, ES, and more)

Usage

swift
let model = try await Qwen3ASRModel.fromPretrained(
    modelId: "aufklarer/Qwen3-ASR-1.7B-MLX-4bit"
)
let text = model.transcribe(audio: samples, sampleRate: 16000)
bash
audio transcribe audio.wav --model aufklarer/Qwen3-ASR-1.7B-MLX-4bit

Variants

VariantSizeModel ID
4-bit~2.1 GBaufklarer/Qwen3-ASR-1.7B-MLX-4bit
8-bit~3.2 GBaufklarer/Qwen3-ASR-1.7B-MLX-8bit
0.6B 4-bit~680 MBaufklarer/Qwen3-ASR-0.6B-MLX-4bit
0.6B 8-bit~1.0 GBaufklarer/Qwen3-ASR-0.6B-MLX-8bit

Links