---
title: MOSS-Audio-Tokenizer-v2-bf16
canonical_url: "https://www.modelscope.cn/models/mlx-community/MOSS-Audio-Tokenizer-v2-bf16"
md_url: "https://www.modelscope.cn/models/mlx-community/MOSS-Audio-Tokenizer-v2-bf16.md"
repository: mlx-community/MOSS-Audio-Tokenizer-v2-bf16
last_updated: 2026-09-03
license: apache-2.0
model_type:
  - moss-audio-tokenizer
architectures:
  - MossAudioTokenizerModel
base_model:
  - OpenMOSS-Team/MOSS-Audio-Tokenizer-v2
base_model_relation: finetune
parameters: 2.1B
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 35
stars: 0
tags:
  - mlx
  - mlx-audio
  - audio
  - codec
  - neural-audio-codec
  - apple-silicon
---

# MOSS-Audio-Tokenizer-v2-bf16

> MOSS-Audio-Tokenizer-v2-bf16 - mlx-community 在 ModelScope 开源的模型。mlx-community/MOSS-Audio-Tokenizer-v2-bf16

mlx-community/MOSS-Audio-Tokenizer-v2-bf16 是 ModelScope 魔搭社区上的 2.1B 参数机器学习模型，采用 apache-2.0 许可，基于 OpenMOSS-Team/MOSS-Audio-Tokenizer-v2 构建。

- **Repository**: mlx-community/MOSS-Audio-Tokenizer-v2-bf16
- **License**: apache-2.0
- **Parameters**: 2.1B
- **Base model**: OpenMOSS-Team/MOSS-Audio-Tokenizer-v2
- **Tags**: mlx, mlx-audio, audio, codec, neural-audio-codec, apple-silicon
- **Downloads**: 35
- **Stars**: 0
- **Last updated**: 2026-09-03

Source: https://www.modelscope.cn/models/mlx-community/MOSS-Audio-Tokenizer-v2-bf16

---

# mlx-community/MOSS-Audio-Tokenizer-v2-bf16

MLX conversion of [OpenMOSS-Team/MOSS-Audio-Tokenizer-v2](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-Tokenizer-v2)
(OpenMOSS) — the 48 kHz stereo neural audio codec behind MOSS-TTS-Local-Transformer-v1.5:
causal windowed transformer encoder/decoder with a residual LFQ quantizer (32 codebooks × 1024,
12.5 frames/s, `downsample_rate` 3840 per channel).

Precision: **bf16** (the upstream `compute_dtype`; the upstream file is fp32, 8.5 GB → 4.3 GB
here). Single `model.safetensors` with the key names mlx-audio's `MossAudioTokenizer` loads
directly (`linear1/2 → ffn.0/2`, `in_projs.0 → in_proj`, …); `config.json` is the upstream
config minus `auto_map`. Round-trip verified with mlx-audio (encode → decode of a 10 s clip).

## Use with mlx-audio

```python
from mlx_audio.codec.models.moss_audio_tokenizer import MossAudioTokenizer
tok = MossAudioTokenizer.from_pretrained("mlx-community/MOSS-Audio-Tokenizer-v2-bf16")
codes = tok.encode_audio("speech.wav", num_quantizers=12)     # [frames, nq] int32
audio = tok.decode_audio_codes(codes, num_quantizers=12)       # [samples, 2] float32 @ 48 kHz
```

Pass `audio_tokenizer_source="mlx-community/MOSS-Audio-Tokenizer-v2-bf16"` to
`MOSS-TTS-Local-Transformer-v1.5` generation to use this mirror instead of the upstream fp32 repo.

## License

Apache-2.0 (upstream weights and this conversion). Please credit OpenMOSS for the model.