Meta Releases Weights for Muse Glimmer 30B Model
Mark Zuckerberg posted that Meta opened weights for a local model with another release planned soon.
Mark Zuckerberg announced Meta is opening the weights for Muse Glimmer, a 30B parameter dense model that runs locally. He added that weights for Muse Spark 1.2, the company's latest foundation model, will follow soon. Alexandr Wang, Chief AI Officer at Meta, posted similar details about the releases and noted the 30B model runs on a single consumer GPU. Multiple replies on X highlighted the Apache 2.0 license and availability through Ollama for local use on Apple Silicon.
Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats…
Reactions
177 postsAW 
Alexandr Wang@alexandr_wangexcited to be releasing open weights for muse glimmer today, a 30b model that runs on a single consumer gpu, with open weights for a version of muse spark 1.2 coming soon. two very different models, both headed into people's hands, with more to come.
- likes: 3.5K
- replies: 126
- bookmarks: 267
- reposts: 211
AW Alexandr Wang@alexandr_wang2/ just like much larger models, muse glimmer can operate as a fully capable agent via planning, tool calls, checking its own results, and failure recovery.
- likes: 663
- replies: 4
- bookmarks: 46
- reposts: 21
AW Alexandr Wang@alexandr_wang3/ muse glimmer was developed with its own architecture and recipe, optimized for its size and agentic performance requirements.
- likes: 626
- replies: 12
- bookmarks: 49
- reposts: 27
AW Alexandr Wang@alexandr_wang4/ to achieve maximum memory efficiency, we quantize model weights to ~4-bit, getting the language model under 20GB with room for the kv cache, perception encoder, and drafter alongside it. a dflash drafter proposes blocks of tokens the main model verifies in parallel, so it…
- likes: 469
- replies: 4
- bookmarks: 33
- reposts: 12
AW Alexandr Wang@alexandr_wang5/ weights on hugging face now. running this week through ollama, LM Studio, vllm, sglang, together, fireworks, and openrouter, with llama.cpp, MLX, and executorch. evaluated under meta's advanced AI scaling framework and cleared for open-weight release.
- likes: 452
- replies: 11
- bookmarks: 24
- reposts: 11
AW Alexandr Wang@alexandr_wang6/ that's the short version. the full write-up, including evals and the training details, is here:
- likes: 503
- replies: 13
- bookmarks: 144
- reposts: 20
