RedHatAI/Muse-Glimmer-30B-FP8-block

This model is a quantized version of meta-models/Muse-Glimmer-30B.

Model Optimizations

This model was obtained by quantizing the weights and activations of meta-models/Muse-Glimmer-30B to FP8 data type, ready for inference with vLLM. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.

Weights are quantized using block-wise FP8 scaling (128×128 blocks), and activations are quantized dynamically per group (group_size=128). Only the weights and activations of the linear operators within transformer blocks are quantized using LLM Compressor. Vision tower, embedding, and output head layers are kept in their original precision.

Creation

from llmcompressor import model_free_ptq

MODEL_ID = "meta-models/Muse-Glimmer-30B"
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-block"

model_free_ptq(
    model_stub=MODEL_ID,
    save_directory=SAVE_DIR,
    scheme="FP8_BLOCK",
    ignore=["re:.*vision.*", "lm_head", "re:.*embed_tokens.*"],
    max_workers=15,
    device="cuda:0",
)

Deployment

docker run --gpus all \
    --privileged --ipc=host -p 8000:8000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    vllm/vllm-openai:muse-glimmer RedHatAI/Muse-Glimmer-30B-FP8-BLOCK \
    --generation-config auto \
    --tensor-parallel-size 1 \
    --enable-auto-tool-choice \
    --tool-call-parser muse_glimmer \
    --reasoning-parser muse_glimmer

For detailed instructions including multi-GPU deployment, multimodal inference, etc see the Muse-Glimmer 30B vLLM usage guide.