Run Muse Glimmer 30B Locally with llama.cpp and DFlash
Deploy Meta's open-weight 30B agentic model on a single consumer GPU using llama.cpp with DFlash speculative decoding for 2-3x throughput improvement.
Meta's Muse Glimmer 30B is an Apache 2.0-licensed open-weight model that runs on a single consumer GPU with 24GB+ VRAM. This recipe deploys it using llama.cpp with DFlash speculative decoding — a block-diffusion drafter that proposes 16 tokens at a time for parallel verification, delivering 2-3x throughput improvement over standard autoregressive generation.
- GPU: 24GB+ VRAM (RTX 4090, RTX 5090, or Mac with 32GB+ unified memory)
- llama.cpp: Latest build with DFlash support (shipped day-0 with Muse Glimmer release)
- Disk space: ~17GB for the smallest GGUF quantization
- OS: Linux, macOS, or Windows (WSL2)
- DFlash with greedy decoding: DFlash works best with deterministic (greedy) decoding, which maximizes speculative token acceptance. Sampling may reduce the speed advantage.
- Sliding-window attention: 39 of 52 layers use 2,048-token sliding windows, keeping long-context inference tractable. Only 13 layers perform global attention.
- Context window: 131,072 tokens stated. For long-context tasks, monitor KV cache memory usage.
- OOM at 24GB: Use the smaller K-quant variant or reduce
--n-gpu-layersto offload some layers to CPU - DFlash not accelerating: Ensure you're using greedy decoding (
--temp 0ordo_sample: false). Sampling increases variance and reduces speculative token acceptance. - Slow on Mac: Ensure Metal acceleration is enabled. Unified memory architectures benefit from the sliding-window attention pattern.
- Meta Research blog: "Introducing Muse Glimmer" (August 10, 2026)
- Hugging Face model card:
meta-models/Muse-Glimmer-30B - DFlash paper: arXiv 2602.06036
Steps
# Clone and build llama.cpp with CUDA support
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
cmake -B build -DGGML_CUDA=ON
cmake --build build --config Release -j $(nproc)
For macOS with Metal:
cmake -B build -DGGML_METAL=ON
cmake --build build --config Release -j $(sysctl -n hw.ncpu)
# Download the official K-quant GGUF from Hugging Face
huggingface-cli download meta-models/Muse-Glimmer-30B-GGUF \
--local-dir ./models/muse-glimmer \
--include "*.gguf"
The smallest quantization is approximately 16.76GB. Total working memory (weights + KV cache + vision encoder + DFlash drafter) targets 24GB.
# Start the llama.cpp server with DFlash speculative decoding
llama serve \
-hf meta-models/Muse-Glimmer-30B-GGUF \
--spec-type draft-dflash \
--spec-draft-n-max 15
The --spec-draft-n-max 15 argument controls how many future tokens DFlash proposes per step. Muse Glimmer's DFlash model was trained with a block size of 16 (one anchor token plus 15 proposed tokens), so values above 15 are clamped.
# Interactive CLI with DFlash
llama cli \
-hf meta-models/Muse-Glimmer-30B-GGUF \
--spec-type draft-dflash
For production deployments with higher throughput:
pip install vllm transformers
# vLLM supports Muse Glimmer via the transformers backend
python -m vllm.entrypoints.openai.api_server \
--model meta-models/Muse-Glimmer-30B \
--backend transformers \
--port 8000
Point your agent at the local endpoint:
# Example: Hermes Agent configuration
export OPENAI_API_BASE=http://localhost:8000/v1
export MODEL_NAME=muse-glimmer-30b
Recipe verified 2026-08-12. Commands are tested but your environment may differ.
Browse related services