Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

MiniMax Sparse Attention

AI overview

Sign in and the AI will write an overview from our coverage.

Headlines · 3

  1. MiniMax releases MXFP8-quantized MiniMax-M3

    MiniMax’s model card lists MiniMax-M3-MXFP8 as an MXFP8-quantized variant of MiniMax-M3, tagged for image-text-to-text tasks. It describes M3 as a native multimodal model with a 1M-token context, about 428B total parameters and about 23B activated parameters, trained from the first step across text, image and video. The model uses MiniMax Sparse Attention, which MiniMax says is designed for million-token contexts and reduces attention compute and memory use compared with GQA. MiniMax also reports 9× prefill and 15× decode speedups over M2 at a 1M-token context, with per-token compute reduced to one-twentieth; these figures are the company’s claims.

    MiniMax · Hugging Face · 🔥 0
  2. MiniMax releases M3 native multimodal model with 1M-token context

    MiniMax published MiniMax-M3 on Hugging Face, describing it as a native multimodal model with a 1M-token context, approximately 428B total parameters and 23B activated parameters. The model card says mixed-modality training combines text, image and video semantics, while MiniMax Sparse Attention targets more efficient million-token inference. MiniMax also claims 9× prefill and 15× decode speedups over M2 at 1M context, but these performance and efficiency figures have not been independently verified.

    MiniMax · Hugging Face · 🔥 0
  3. vLLM Adds Day-0 Support for MiniMax-M3.1-Flash

    vLLM Blog reports day-0 support for MiniMax-M3.1-Flash on NVIDIA Blackwell GPUs, NVIDIA Rubin GPUs with dedicated kernels, and AMD MI350-series GPUs. Prebuilt NVIDIA and AMD images are available. The implementation uses MiniMax Sparse Attention, an NVFP4 KV cache, NVFP4-quantized MoE weights and a DSpark drafter. vLLM reports gains in kernel and serving performance, including optimized agentic serving on GB300, Rubin and MI355X hardware. It also reports day-0 end-to-end reinforcement-learning support with veRL, FSDP2 and LoRA, demonstrated on 16 GB300 GPUs.

    vLLM Blog · 🔥 0

Experience and discussion from the community

Share my MiniMax Sparse Attention experienceAsk about MiniMax Sparse Attention

Nobody has shared their experience with MiniMax Sparse Attention yet.