hackernews.pink

Top Stories Trending New For You Bookmarks Pink Dossier

hackernews.pink

1209 points•639 comments•3w ago•by riordan•research.meta.ai ↗

Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

Meta unveiled Muse Glimmer, a 30-billion-parameter open-weights model for on-device agent workflows, released under Apache 2.0 with weights on Hugging Face. It’s optimized for local deployment on consumer hardware, using 4-bit quantization to fit under about 20 GB and a DFlash-based drafter to speed generation, with integrations planned for llama.cpp, MLX, and ExecuTorch. Muse Glimmer supports end-to-end agent tasks, tool use, multi-step reasoning, multimodal input, and multilingual data, and has been benchmarked against similar-size models with competitive results.

Discussion (639)

No comments yet

Top
Trending
New
For You
Bookmarks
Dossier