MiniMax-M2.5-JANG_3L

<p align="center"> <a href="https://mlx.studio"><img src="https://raw.githubusercontent.com/jjang-ai/jangq/main/assets/mlx-studio-light.png" alt="MLX Studio" width="500"></a> </p> <h4 align="center"><a href="https://mlx.studio">MLX Studio</a> — native JANG support with reasoning</h4>

---

<p align="center"> <img src="https://raw.githubusercontent.com/jjang-ai/jangq/main/assets/jangq-logo-dark.png" alt="JANG" width="300"> </p> <p align="center"> <a href="https://vmlx.net"><img src="vmlx-app.png" alt="vMLX — run JANG models on Apple Silicon" width="820"></a> </p> <h3 align="center">⚡ All JANG models are meant to be run in <a href="https://vmlx.net">vMLX</a></h3> <h3 align="center">MiniMax M2.5 (227B-A21B) — JANG_3L (3.08-bit) — Reasoning</h3> <p align="center"><b>JANG</b> — Jang Adaptive N-bit Grading | The GGUF Equivalent for MLX</p> <p align="center"> <a href="https://github.com/jjang-ai/jangq"><img src="https://img.shields.io/badge/GitHub-Source_Code-blue?logo=github" alt="GitHub"></a>&nbsp; <a href="https://pypi.org/project/jang/"><img src="https://img.shields.io/pypi/v/jang?label=PyPI&color=green" alt="PyPI"></a>&nbsp; <a href="https://jangq.ai"><img src="https://img.shields.io/badge/Web-jangq.ai-orange" alt="Website"></a>&nbsp; <a href="https://x.com/dealignai"><img src="https://img.shields.io/badge/X-@dealignai-black?logo=x" alt="X/Twitter"></a> </p>

JANG is fully open-source. Quantization engine and full commit history: github.com/jjang-ai/jangq. Created by Jinho Jang.

---

Supported apps: MLX Studio (full native support) and oMLX (PR #364). LM Studio, Ollama, and Inferencer do not yet support JANG.

---

Why JANG models?

Tools like mlx-lm, oMLX (oQ), and others can quantize models — but shipping a tool is the easy part. JANG models come from hundreds of hours of per-architecture testing: finding which layers break at which bit depths, which MoE routing survives quantization, which models need bfloat16 to avoid NaN. We don't just quantize — we convert, verify, benchmark, and publish every model with tested scores. No other project in the MLX ecosystem publishes pre-tested quantized models at this scale.

---

Results: 93.5% MMLU (200 Questions, Smart Two-Pass)

Pass 1 (no-thinking): 158/200 (79.0%) | Pass 2 (reasoning retry): +29 recovered

JANG vs MLX — MiniMax M2.5

MLX is broken on MiniMax at ALL bit levels (~25% = random chance). JANG is the ONLY working quantization for MiniMax M2.5 on Apple Silicon.

Key Features

• 93.5% MMLU — five subjects at 100%
• 41 tok/s generation on M3 Ultra
• 82 GB on disk — fits 96+ GB Macs
• 227B total / 21B active — 256 MoE experts, top-8 routing
• Reasoning mode: <think>...</think> step-by-step reasoning
• Sigmoid + bias routing: MiniMax-specific MoE (not softmax)
• FP8 source with block-wise 128x128 scales

Important Notes

• Temperature must be 1.0 — greedy decoding (temp=0) causes infinite thinking loops on MiniMax
• Tokenizer: Known-good tokenizer included (mlx_lm.convert corrupts MiniMax tokenizer)

Architecture

227B total parameters, 21B active per token
- 64 layers, all MoE (256 experts, top-8 routing)
- Sigmoid + bias expert routing (non-normalized)
- GQA attention: 48 heads, 8 KV heads
- FP8 E4M3 source with block-wise scales

Install

pip install jang[mlx]

---

Created by Jinho Jang — jangq.ai — @dealignai