# MiniMax M3 MXFP8

MiniMax M3 MXFP8, MiniMax का AI मॉडल है। इसके वेट सार्वजनिक हैं। इसे पहली बार 2026-10-07 को दर्ज किया गया; रिलीज़ की तारीख की अभी पुष्टि नहीं हुई है।

- डेवलपर: MiniMax
- Id: MiniMaxAI/MiniMax-M3-MXFP8
- रिपोर्टों के अनुसार: 2026-08-26T12:57:44.000Z
- खोज: 2026-10-07T19:40:22.731Z
  - [संदर्भ लिंक](https://huggingface.co/MiniMaxAI/MiniMax-M3-MXFP8)
- स्थिति: जारी
- वेट्स का संस्करण: 2026-06-12
  - [संदर्भ लिंक](https://huggingface.co/api/models/MiniMaxAI/MiniMax-M3-MXFP8/paths-info/d2be410f483297907301ec94fdf0ca5385f9b169)
- कैनोनिकल: https://alexgetman.com/hi/models/minimax/minimax-m3-mxfp8/

## प्रश्न

### क्या MiniMax M3 MXFP8 के वेट खुले हैं?

हाँ। MiniMax वेट प्रकाशित करता है।

### MiniMax M3 MXFP8 कब जारी हुआ?

रिलीज़ की तारीख की अभी पुष्टि नहीं हुई है। Signal Forge ने इसे पहली बार 2026-10-07 को दर्ज किया।

## समयरेखा

- 2026-11-13T19:04:38.386Z · [GitHub search · MiniMax · pulls · ROCm/ATOM · @itej89](https://github.com/ROCm/ATOM/pull/2508)
- 2026-11-13T19:04:38.386Z · [GitHub search · MiniMax · pulls · inference-sim/blis-vllm · @sriumcp](https://github.com/inference-sim/blis-vllm/pull/131)
> own `ModelConfig` over the catalog `config.json` of `deepseek-v4-pro`, `minimax-m3-mxfp8` and `minimax-m3-nvfp4`, for every method name vLLM lists, none, and a name it does not list;
- 2026-11-13T19:05:09.193Z · [GitHub search · MiniMax · commits · rodrigueztaylor87/rod · @rodrigueztaylor87](https://github.com/rodrigueztaylor87/rod/commit/64189a80753eb37b84793d886ded01ab340508fb)
> 本文基于GPUStack平台，使用vLLM专用镜像完成MiniMax-M3-MXFP8模型在四卡H20-3e上的部署，支持张量并行、EAGLE3投机解码及多模态推理。实测显示开启EAGLE3后推理速度从126 tokens/s提升至200 tokens/s，加速近60%。涵盖环境配置、镜像拉取、自定义vLLM版本注册、启动参数调优（含mm-encoder-attn-backend、speculative-config等）、chat_template修复及基准测试全流程。
- 2026-11-15T10:42:30.398Z · [GitHub search · MiniMax · commits · ramirezashley0672/ram · @ramirezashley0672](https://github.com/ramirezashley0672/ram/commit/3e865771da594ccf6f2204f0b626cf384481ae70)
> 本文详解MiniMax-M3-MXFP8模型的MXFP8量化技术及其本地部署方法。MXFP8通过将参数压缩至8位格式，在保持性能前提下降低50%内存占用、提升推理速度并减少硬件成本；结合MiniMax Sparse Attention（MSA）技术，显著优化长上下文（1M tokens）下的Attention计算效率。部署涵盖模型下载、SGLang/vLLM/Transformers框架适配及关键推理参数配置。
- 2026-11-15T12:48:39.148Z · [GitHub search · MiniMax · pulls · inference-sim/blis-substrate · @sriumcp](https://github.com/inference-sim/blis-substrate/pull/42)
- 2026-11-15T12:48:39.148Z · [GitHub search · MiniMax · pulls · inference-sim/blis-substrate · @sriumcp](https://github.com/inference-sim/blis-substrate/pull/53)
- 2026-11-15T13:19:24.965Z · [GitHub search · MiniMax · pulls · future-agi/future-agi · @a692570](https://github.com/future-agi/future-agi/pull/3144)
> models: [MiniMaxAI/MiniMax-M3-MXFP8]
- 2026-11-15T13:34:54.185Z · [GitHub search · MiniMax · pulls · vllm-project/vllm · @avininjamay8](https://github.com/vllm-project/vllm/pull/66646)
> The MiniMax-M3-MXFP8 wide-EP cells are very slow. The 2P2D EP16 GSM8K eval takes about 1h37 in CI and the 1P1D EP8 one about an hour.
- 2026-11-15T13:50:16.449Z · [GitHub search · MiniMax · pulls · vllm-project/vllm-ascend · @FORFuture37](https://github.com/vllm-project/vllm-ascend/pull/17764)
> - Model weights: `MiniMax/MiniMax-M3-MXFP8` from ModelScope
- 2026-11-15T13:50:16.449Z · [GitHub search · MiniMax · pulls · vllm-project/vllm-ascend · @FORFuture37](https://github.com/vllm-project/vllm-ascend/pull/19129)
> - Served model `minimax-m3` from `/mnt/share/weight/MiniMax-M3-MXFP8`.
- 2026-11-15T15:09:30.906Z · [GitHub search · MiniMax · pulls · ROCm/ATOM · @kliuae](https://github.com/ROCm/ATOM/pull/2043)
- 2026-11-15T15:09:30.906Z · [GitHub search · MiniMax · pulls · sgl-project/sglang · @waqahmed-amd-fi](https://github.com/sgl-project/sglang/pull/39994)
- 2026-11-15T15:27:30.324Z · [GitHub search · MiniMax · commits · inference-sim/blis-substrate · @sriumcp](https://github.com/inference-sim/blis-substrate/commit/8e6b55aa64bd1e94c2304998e701f091ed84b8c1)
> quantisations that are separate catalog entries (MiniMax-M3 and MiniMax-M3-MXFP8 share the EAGLE3 drafter Inferact/MiniMax-M3-EAGLE3). model.yaml's target,
- 2026-11-15T15:59:52.324Z · [GitHub search · MiniMax · commits · vllm-project/vllm-ascend · @FORFuture37](https://github.com/vllm-project/vllm-ascend/commit/514a647be60f6044bb542f7be2529619559b1a15)
> - Served model `minimax-m3` from `/mnt/share/weight/MiniMax-M3-MXFP8`.
- 2026-11-15T16:29:02.877Z · [GitHub search · MiniMax · pulls · vllm-project/vllm · @Pb314314](https://github.com/vllm-project/vllm/pull/65972)
> (a local patch, not in this PR) with MiniMax-M3-MXFP8,
- 2026-11-15T16:46:23.763Z · [GitHub search · MiniMax · commits · inference-sim/blis-vllm · @sriumcp](https://github.com/inference-sim/blis-vllm/commit/0019434826bc270d129ca37ce30cb9e6d76cc7c0)
> (interfaces.py:339-385, scheduler.py:271-284). The catalog graphs of minimax-m3, minimax-m3-mxfp8 and kimi-k3 hold the
- 2026-11-15T18:07:44.690Z · [GitHub search · MiniMax · pulls · vllm-project/vllm · @hmellor](https://github.com/vllm-project/vllm/pull/61813)

## संदर्भ लिंक

- [GitHub search · MiniMax · pulls · ROCm/ATOM · @kliuae](https://github.com/ROCm/ATOM/pull/2043)
- [GitHub search · MiniMax · pulls · ROCm/ATOM · @itej89](https://github.com/ROCm/ATOM/pull/2508)
- [GitHub search · MiniMax · pulls · future-agi/future-agi · @a692570](https://github.com/future-agi/future-agi/pull/3144)
- [GitHub search · MiniMax · commits · inference-sim/blis-substrate · @sriumcp](https://github.com/inference-sim/blis-substrate/commit/8e6b55aa64bd1e94c2304998e701f091ed84b8c1)
- [GitHub search · MiniMax · pulls · inference-sim/blis-substrate · @sriumcp](https://github.com/inference-sim/blis-substrate/pull/42)
- [GitHub search · MiniMax · pulls · inference-sim/blis-substrate · @sriumcp](https://github.com/inference-sim/blis-substrate/pull/53)
- [GitHub search · MiniMax · commits · inference-sim/blis-vllm · @sriumcp](https://github.com/inference-sim/blis-vllm/commit/0019434826bc270d129ca37ce30cb9e6d76cc7c0)
- [GitHub search · MiniMax · pulls · inference-sim/blis-vllm · @sriumcp](https://github.com/inference-sim/blis-vllm/pull/131)
- [GitHub search · MiniMax · commits · ramirezashley0672/ram · @ramirezashley0672](https://github.com/ramirezashley0672/ram/commit/3e865771da594ccf6f2204f0b626cf384481ae70)
- [GitHub search · MiniMax · commits · rodrigueztaylor87/rod · @rodrigueztaylor87](https://github.com/rodrigueztaylor87/rod/commit/64189a80753eb37b84793d886ded01ab340508fb)
- [GitHub search · MiniMax · pulls · sgl-project/sglang · @waqahmed-amd-fi](https://github.com/sgl-project/sglang/pull/39994)
- [GitHub search · MiniMax · commits · vllm-project/vllm-ascend · @FORFuture37](https://github.com/vllm-project/vllm-ascend/commit/514a647be60f6044bb542f7be2529619559b1a15)
- [GitHub search · MiniMax · pulls · vllm-project/vllm-ascend · @FORFuture37](https://github.com/vllm-project/vllm-ascend/pull/17764)
- [GitHub search · MiniMax · pulls · vllm-project/vllm-ascend · @FORFuture37](https://github.com/vllm-project/vllm-ascend/pull/19129)
- [GitHub search · MiniMax · pulls · vllm-project/vllm · @hmellor](https://github.com/vllm-project/vllm/pull/61813)
- [GitHub search · MiniMax · pulls · vllm-project/vllm · @Pb314314](https://github.com/vllm-project/vllm/pull/65972)
- [GitHub search · MiniMax · pulls · vllm-project/vllm · @avininjamay8](https://github.com/vllm-project/vllm/pull/66646)
- [huggingface:MiniMaxAI](https://huggingface.co/MiniMaxAI/MiniMax-M3-MXFP8)
- [hf-author-artifacts:MiniMaxAI](https://huggingface.co/api/models/MiniMaxAI/MiniMax-M3-MXFP8/paths-info/d2be410f483297907301ec94fdf0ca5385f9b169)

## संकेत