---
title: MOSS-Music-8B-Thinking-6bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/MOSS-Music-8B-Thinking-6bit"
md_url: "https://www.modelscope.cn/models/mlx-community/MOSS-Music-8B-Thinking-6bit.md"
repository: mlx-community/MOSS-Music-8B-Thinking-6bit
last_updated: 2026-06-28
license: apache-2.0
model_type:
  - moss_music
architectures:
  - MossMusicModel
base_model:
  - OpenMOSS-Team/MOSS-Music-8B-Thinking
base_model_relation: finetune
parameters: 2.5B
tensor_type:
  - BF16
  - U32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
  - zh
downloads: 56
stars: 1
tags:
  - mlx
  - audio
  - music
  - music-understanding
  - audio-language-model
  - moss
  - moss-music
---

# MOSS-Music-8B-Thinking-6bit

> MOSS-Music-8B-Thinking-6bit - mlx-community 在 ModelScope 开源的模型。MOSS-Music-8B-Thinking · MLX 6-bit

mlx-community/MOSS-Music-8B-Thinking-6bit 是 ModelScope 魔搭社区上的 2.5B 参数机器学习模型，采用 apache-2.0 许可，基于 OpenMOSS-Team/MOSS-Music-8B-Thinking 构建。

- **Repository**: mlx-community/MOSS-Music-8B-Thinking-6bit
- **License**: apache-2.0
- **Parameters**: 2.5B
- **Base model**: OpenMOSS-Team/MOSS-Music-8B-Thinking
- **Tags**: mlx, audio, music, music-understanding, audio-language-model, moss, moss-music
- **Downloads**: 56
- **Stars**: 1
- **Last updated**: 2026-06-28

Source: https://www.modelscope.cn/models/mlx-community/MOSS-Music-8B-Thinking-6bit

---

# MOSS-Music-8B-Thinking · MLX 6-bit

[![Base model](https://img.shields.io/badge/Base-MOSS--Music--8B--Thinking-ff9e2c?logo=huggingface&logoColor=white)](https://huggingface.co/OpenMOSS-Team/MOSS-Music-8B-Thinking)
[![MLX](https://img.shields.io/badge/MLX-Apple%20Silicon-black)](https://github.com/ml-explore/mlx)
[![License](https://img.shields.io/badge/License-Apache--2.0-green)](https://www.apache.org/licenses/LICENSE-2.0)
[![Backend](https://img.shields.io/badge/Backend-moss__music__mlx-blue?logo=github)](https://github.com/OpenMOSS/MOSS-Music/pull/3)

A **6-bit [MLX](https://github.com/ml-explore/mlx) quantization** of
[OpenMOSS-Team/MOSS-Music-8B-Thinking](https://huggingface.co/OpenMOSS-Team/MOSS-Music-8B-Thinking)
for music understanding on Apple Silicon. A ~8 GB build that stays essentially lossless.

> Community conversion, not an official release. All model credit goes to the OpenMOSS Team.

Other sizes: [8-bit](https://huggingface.co/mlx-community/MOSS-Music-8B-Thinking-8bit) · [4-bit](https://huggingface.co/mlx-community/MOSS-Music-8B-Thinking-4bit)

## Usage

MOSS-Music is a custom multimodal (audio + text) model, so it does **not** load with
`mlx_lm` / `mlx_vlm` directly. Use the `moss_music_mlx` backend
([code](https://github.com/dthinkr/MOSS-Music/tree/feat/mlx-backend/mlx), [PR](https://github.com/OpenMOSS/MOSS-Music/pull/3)):

```python
from huggingface_hub import snapshot_download
from moss_music_mlx import load_pretrained, generate
from src.processing_moss_music import MossMusicProcessor

path = snapshot_download("mlx-community/MOSS-Music-8B-Thinking-6bit")
model = load_pretrained(path)
proc = MossMusicProcessor.from_pretrained(path, trust_remote_code=True, enable_time_marker=True)
print(generate(model, proc, "Analyze this track: genre, key, BPM, structure.", audio_path="song.mp3"))
```

## Conversion

- **6-bit, group size 64.** The audio encoder is kept at **bf16** to preserve audio
  fidelity; quantization is applied to the Qwen3 layers, token embeddings and `lm_head`.
- Converted with `mlx==0.31.2`, `mlx-lm==0.29.1`.

## Accuracy

Versus the fp32 PyTorch reference, the 6-bit model's prefill next-token argmax is
identical and the logits match to cosine **0.99989** (vs 0.99999 for 8-bit), effectively lossless.

## License & credit

Apache-2.0, inherited from the base model. This repository provides only the
MLX-quantized weights; all credit goes to the OpenMOSS Team.