muse-glimmer

Private, on-device AI desktop app — GGUF (llama.cpp) & MLX, running Muse-Glimmer and Qwen3.8 with their native reasoning-effort ladders wired in as real controls. Local coding agent, RAG knowledge base, Deep Research, vision and voice. 100% offline, no account, no telemetry. Windows & macOS.

Introducing Muse Glimmer: open-weight 30B agentic multimodal model that runs on your device (Meta). Interactive local agent lab + guide. Apache 2.0 · on-device AI · function calling

Reproducible RDNA reference for Meta Muse-Glimmer-30B — adapting MI-series ROCm recipes to Ryzen AI and Radeon with vLLM, llama.cpp, DFlash and auditable benchmarks.

Open-source music-business contract agent built on Meta's Muse Glimmer 30B. Extracts deal terms, flags red-flag clauses, reads scanned PDFs.

Automatic overview + tiles HD vision and OCR proxy for Muse Glimmer 30B

Measured Muse Glimmer 30B recipe for NVIDIA DGX Spark, with llama.cpp, DFlash parity checks, vision, tools, and GB10 benchmarks.

PKGBUILD bump of AUR llama.cpp-hip to latest llama.cpp (b10362) for Muse Glimmer support

MLX-native Muse Glimmer 30B QLoRA parody for Apple Silicon. High-performance inference for Zuck-class workloads.

Muse-Glimmer-30B NVFP4 + DFlash on NVIDIA DGX Spark (GB10): 27.5 tok/s avg, 90/100 tool-eval, 128K context.

Local Muse Glimmer 30B vision-and-reasoning playground for high-memory Apple silicon Macs.

Telegram-бот на моделях NVIDIA API: стриминг ответов, веб-поиск с проверкой источников и публикацией длинных ответов на Telegraph, roadmap по стеку и техсобеседование по Go