Officially

From the publisher, with its history intact.

Hugging Face huggingface.co

Meta is back with Muse Glimmer: local, agentic, multimodal, and open source

Imported from official source

Research

AI Classified by Officially

To celebrate, we are shipping with Meta day-0 support in transformers, llama.cpp, vLLM, Inference Endpoints, and other libraries. We built a few cool things and explain our findings in this blog.

Scores are reported as published. Bold indicates the best result among the compared models; ↓ indicates lower is better.

Muse Glimmer is a dense 30B parameter model consisting of:

  • 2B ViT-style encoder for vision (Perception Encoder)

    • In addition to the main VLM, there’s also a speculative decoding drafter implemented on DFlash. Usage of this module is optional, and it can provide much faster generation in exchange for some memory cost. We found this drafter to be particularly well suited to structured content generation such as coding.

    The language model uses the following architecture components:

  • Hybrid attention: Alternating between three sliding window layers (of 2,048 tokens) using rotary position embedding, followed by a fourth layer that uses full attention and NoPE (no positional embedding). The pattern is therefore (SWA, SWA, SWA, Full), repeated 13 times to a total of 52 layers. This allows the model to retain relative order and distance information with RoPE and preserve information globally with NoPE.
  • Gated Grouped-Query Attention: Each key-value head is shared by 16 query heads, which reduces KV-cache memory by 16x and makes generation faster and cheaper.
  • Q-K normalization with extra query scaling: Before computing attention, Muse Glimmer applies RMS normalization to every query and key head to keep attention logits stable. After this, queries are multiplied by a scale factor to set the target logit scale after normalization. The extra query scaling behaves like an inverse temperature at the softmax level.

    • This is an extract. The publication continues at the source.

    Read the original at the source: https://huggingface.co/blog/muse-glimmer

    Officially imported this from Hugging Face’s own source and shows an extract. If you work there, claiming the profile and verifying the domain lets you choose to show the full text here.

    Provenance

    Organization
    Hugging Face — imported from official source
    Official source
    https://huggingface.co/blog/feed.xml RSS
    Imported
    September 18, 2026 09:42
    Versions
    1 recorded
    Identity
    https://huggingface.co/blog/muse-glimmer

    Officially records where a publication came from, not whether it is true. Imported records are reproduced from an organization's own official source.