Muse-Glimmer-30B-NVFP4

About

NVIDIA's FP4-quantized release of Meta Superintelligence Lab's Muse-Glimmer-30B, a 29.6B-parameter dense causal transformer with a perception encoder that accepts interleaved text and images. Quantizing the weights with NVIDIA Model Optimizer lowers the precision to NVFP4, which reduces the memory footprint so the model can run for agentic work (multi-step planning, tool use, coding) on consumer hardware rather than server GPUs. It handles context up to 131K tokens and exposes four reasoning-strength settings to trade output quality against speed.