MiniMax-M3-AWQ-INT4

FenomAI/MiniMax-M3-AWQ-INT4

MiniMax-M3-AWQ-INT4 is a image-text-to-text model from FenomAI. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as other.

<div align="center" <img src="https://huggingface.co/buckets/cyankiwi/activation-aware-2.0/resolve/banner/cyankiwi-banner-awq-0.png" </div

Downloads · 30 days

17% of all-time downloads

All-time downloads

Public

Parameters

453B

510 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors258 GB · 100%

Parameter types

How the weights are stored.

I32432B · 95%

Try a prompt

Base models

MiniMaxAI/MiniMax-M3

Model card

From the Hugging Face model README

<div align="center"> <img src="https://huggingface.co/buckets/cyankiwi/activation-aware-2.0/resolve/banner/cyankiwi-banner-awq-0.png"> </div> <div align="left"> <table align="center" style="border-collapse:collapse; border:none;"> <tr style="border:none;"> <td align="right" style="border:none; padding:4px 12px 4px 0;"><b>Version</b></td> <td align="left" style="border:none; padding:4px 0;">26.05.01</td> </tr> <tr style="border:none;"> <td align="right" style="border:none; padding:4px 12px 4px 0;"><b>Calibration</b></td> <td align="left" style="border:none; padding:4px 0;"> <a href="https://huggingface.co/datasets/cyankiwi/calibration" target="_blank">STEM and Agentic</a> </td> </tr> <tr style="border:none;"> <td align="right" style="border:none; padding:4px 12px 4px 0;"><b>Languages</b></td> <td align="left" style="border:none; padding:4px 0;"> <code>EN</code> <code>ZH</code> <code>HI</code> <code>AR</code> <code>RU</code> <code>JA</code> <code>KO</code> <code>NL</code> <code>FR</code> <code>ES</code> </td> </tr> <tr style="border:none;"> <td align="right" style="border:none; padding:4px 12px 4px 0;"><b>Model Size</b></td> <td align="left" style="border:none; padding:4px 0;">240.30 GB</td> </tr> <tr style="border:none;"> <td align="right" style="border:none; padding:4px 12px 4px 0;"><b>Contact</b></td> <td align="left" style="border:none; padding:4px 0;"> <a href="mailto:ton@cyan.kiwi">Email</a> </td> </tr> </table> </div>

Serving with vLLM

This checkpoint needs a patched vLLM (MiniMax-M3 compressed-tensors support). The patch is Python-only, so it installs on top of upstream's precompiled binaries — no CUDA compilation.

Install

# uv (skip if already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# clone the fork + fetch the upstream base commit
git clone https://github.com/toncao/vllm.git
cd vllm
git remote add upstream https://github.com/vllm-project/vllm.git
git fetch upstream a7fdfeef72323eb3db6f0620e4ea200290d0ca5a
git checkout minimax-m3-compressed-tensors

# Python 3.12 env + install with upstream precompiled kernels
uv venv --python 3.12
source .venv/bin/activate
VLLM_USE_PRECOMPILED=1 uv pip install -e . --torch-backend=auto

Serve

vllm serve cyankiwi/MiniMax-M3-AWQ-INT4 --block-size 128

MiniMax-M3 is a native multimodal model with 1M context. It has ~428B parameters and ~23B activated parameters.

Highlights:

Native Multimodality: M3 undergoes mixed-modality training from the very first step, enabling deeper semantic fusion across text, image, and video.
Context Scaling via Sparse Attention: M3 introduces MiniMax Sparse Attention (MSA) to improve long context efficiency. M3 delivers 9× prefill and 15× decode speedups compared to M2 at 1M context, reducing per-token compute to 1/20.
Coding & Cowork Capability: M3 achieves frontier-level performance across long-horizon agentic benchmarks, excelling in both coding and cowork.

MiniMax Sparse Attention (MSA)

M3 is powered by MiniMax Sparse Attention (MSA), a high-performance sparse attention operator designed for million-token contexts. Compared with GQA, MSA dramatically reduces the attention compute and memory footprint while preserving model quality.

📄 Read the technical report: arXiv:2606.13392 · Hugging Face Papers

How to Use

M3 supports two reasoning modes:

thinking — for complex reasoning, agentic tasks, and long-horizon collaboration.
non-thinking — for latency-sensitive scenarios such as chat and code completion.

Local Deployment

Download the model:

hf download MiniMaxAI/MiniMax-M3 --local-dir MiniMax-M3

We recommend the following inference frameworks (listed alphabetically) to serve the model:

Inference Parameters

We recommend the following parameters for best performance: temperature=1.0, top_p=0.95, top_k=40.

Contact Us

Task: Image-Text-to-Text
Library: transformers
Type: minimax_m3_vl
License: other
Created: Jun 30, 2026
Updated: Jun 30, 2026