vmlinux/Muse-Glimmer-30B-ROCmFPX-GGUF · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
AI Model Release
Summary
Meta Superintelligence Labs has introduced Muse Glimmer, a 30B-parameter open agentic model designed specifically for local execution on consumer hardware. The model utilizes logit distillation from larger models to enable advanced capabilities like function calling and coding without requiring cloud infrastructure.
Highlights
- Optimized for always-on local workflows on Mac or PC with a single consumer GPU
- Released as open weights under the permissive Apache 2.0 license
- Capable of complex tasks including long-horizon execution, tool calling, and multimodal understanding
- Trained using a novel distillation recipe from the larger Muse Spark model
- Planned integrations include llama.cpp, MLX, and ExecutorTorch
auto-generated
Context
Audience
Developers, AI Researchers, and Machine Learning Engineers
DomainMachine Learning
Formatlong-form article
Accessopen source
Source Overview
Muse Glimmer is a 30B-parameter agentic model from Meta Superintelligence Labs, released under Apache 2.0 and sized to run on a single consumer GPU.
It handles local agent workflows, function calling, coding, and multimodal input, trained via distillation from a larger teacher model. The model uses 4-bit quantization and speculative decoding to fit within 24-32 GB of VRAM, with integrations for llama.cpp, MLX, and ExecuTorch.
Topics
Related
Hugging Facellama.cppMLXExecuTorchLLM-as-a-judgeLogit Distillation
