---
title: LFM3-Audio-2B-JANG_6M
canonical_url: "https://www.modelscope.cn/models/OsaurusAI/LFM3-Audio-2B-JANG_6M"
md_url: "https://www.modelscope.cn/models/OsaurusAI/LFM3-Audio-2B-JANG_6M.md"
repository: OsaurusAI/LFM3-Audio-2B-JANG_6M
last_updated: 2026-10-28
license: other
model_type:
  - lfm3_audio
architectures:
  - Lfm3AudioForConditionalGeneration
base_model:
  - LiquidAI/LFM3-Audio-2B
base_model_relation: quantized
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
  - ar
  - zh
  - fr
  - de
  - it
  - ja
  - ko
  - pt
  - es
  - vi
  - th
  - id
  - hi
  - ru
  - pl
downloads: 8
stars: 0
tags:
  - mlx
  - jang
  - jang_6m
  - quantized
  - apple-silicon
  - lfm3
  - liquid
  - osaurus
---

# LFM3-Audio-2B-JANG_6M

> LFM3-Audio-2B-JANG_6M - OsaurusAI 在 ModelScope 开源的模型。OsaurusAI/LFM3-Audio-2B-JANG6M

OsaurusAI/LFM3-Audio-2B-JANG_6M 是 ModelScope 魔搭社区上的机器学习模型，采用 other 许可，基于 LiquidAI/LFM3-Audio-2B 构建。

- **Repository**: OsaurusAI/LFM3-Audio-2B-JANG_6M
- **License**: other
- **Base model**: LiquidAI/LFM3-Audio-2B
- **Tags**: mlx, jang, jang_6m, quantized, apple-silicon, lfm3, liquid, osaurus
- **Downloads**: 8
- **Stars**: 0
- **Last updated**: 2026-10-28

Source: https://www.modelscope.cn/models/OsaurusAI/LFM3-Audio-2B-JANG_6M

---

<p align="center"><a href="https://osaurus.ai"><img src="./osaurus-x-banner.png" alt="Osaurus AI"></a></p>

# OsaurusAI/LFM3-Audio-2B-JANG_6M

JANG_6M build of [LiquidAI/LFM3-Audio-2B](https://huggingface.co/LiquidAI/LFM3-Audio-2B) for Apple Silicon. LFM3-Audio 2B: the LFM3 trunk with a 17-layer LFM2 speech encoder (16 kHz).

8-bit token mixers, 6-bit FFN, and every quantized matrix calibrated with GPTQ on the exact
storage grid plus activation-aware scaling, using calibration data weighted toward tool calls,
terminal agents, reasoning and code.

## Bundle

| | |
|---|---|
| Modalities | text, speech in |
| On-disk size | 2.01 GiB (one shard), float16 |
| Attention q/k/v/o, short-conv in/out projections | 8-bit affine, group 32, GPTQ |
| FFN w1 / w3 / w2 | 6-bit affine, group 32, AWQ + GPTQ |
| Tied embedding / lm_head | 8-bit affine, group 32 (it is the output head) |
| Kept at full precision | attention output gate, depthwise conv kernels, all norms, speech encoder and projector, speech-output decoder (unused for text) |
| GPTQ | 142/142 tensors: fixed min-max grid (byte-identical layout to `mx.quantize`), act-order, Hessians accumulated over every calibration token, w2 Hessian rebuilt through the already-quantized w1/w3, best-of-RTN guard; mean calibration reconstruction error -53.5% vs round-to-nearest |
| AWQ | channel scales (alpha 0.25, clip 0.5-2) folded into `ffn_norm` and `w3` rows; function-preserving, no runtime cost |

## Calibration (529,024 tokens)

Every sample is rendered through the model's own chat template: structured `tools`, assistant
tool calls in LFM3's `<tool_call><function=...>` form, tool results as `tool` turns, and agent
thoughts inside `<think>`. Part of the set is the model's own answers (card sampling) to
reasoning, tool, terminal and coding prompts, so its serving distribution is in calibration.

| domain | tokens |
|---|---|
| tool calling (Hermes, native XML calls) | 154,969 |
| terminal / SWE agent trajectories | 174,363 |
| reasoning (long <think> traces) | 105,223 |
| coding and agentic text | 46,072 |
| general, science, systems, security, multilingual, long context | 37,598 |
| speech (ASR, spoken QA, spoken instructions) | 10,799 |

## Measured against the bf16 source

64 held-out sequences never seen in calibration (13,958 scored
positions: tool calls, terminal and SWE agents, reasoning, code, general text, speech).
KL divergence in nats per position (lower is better); top-1 is argmax agreement with bf16.

| build | size | KL median | KL mean | KL p95 | top-1 |
|---|---|---|---|---|---|
| **this bundle** | 2.01 GiB | 0.00004 | 0.00119 | 0.0049 | 98.95% |
| same layout, round-to-nearest | 2.01 GiB | 0.00006 | 0.00190 | 0.0062 | 98.76% |
| uniform 6-bit gs64 (mlx-vlm default recipe) | 1.77 GiB | 0.00016 | 0.00424 | 0.0171 | 97.97% |

## Use

- Runtimes with LFM3 support: vMLX / Osaurus (vmlx-swift with LFM3, Osaurus with LFM3 media
  support), and mlx-vlm with the LFM3 port.
- Thinking: no thinking mode. Tool calls: XML function format. Sampling defaults: temperature 0.0, top_k 1, repetition_penalty 1.0.
- License: LFM Open License v1.0 (see `LICENSE`), same as the base model.