ginfer

Muse-Glimmer-30B, GInfer NVFP4

This repo holds the GInfer packages of Muse-Glimmer-30B in NVFP4: single-file packages for the GInfer inference engine (and GChat, which is built on it). They run only on GInfer / GChat. They are not safetensors or GGUF files and other runtimes cannot load them.

GInfer is Sectile Labs' inference engine for fast, high-concurrency, long-context work on NVIDIA GPUs.

Packages

File Drafter Tensor-parallel size Size in bytes SHA-256
muse_glimmer_30b_nvfp4_df2q4.ginfer Q4 DFlash2 1 (one GPU) 22,360,828,416 878d6de4e0742f07fd7e10ca1efd819111a894325435d0297bf352ef72ae8d5c
muse_glimmer_30b_nvfp4_df2q4_tp2.ginfer Q4 DFlash2 2 (two GPUs) 22,361,054,208 89bd89337cae7f74b90f737c662e2552c3b14d6af617b0a0a495f40f3535a978
muse_glimmer_30b_nvfp4_df2.ginfer NVFP4 DFlash2 1 (one GPU) 22,447,498,752 6d6cc348c75a75c60ca011d267eb389dcbb8530f6c51a680dad911d25365b61f
muse_glimmer_30b_nvfp4_df2_tp2.ginfer NVFP4 DFlash2 2 (two GPUs) 22,447,761,668 38e0d3de4c5c7371c88c53bd6ed4a7145c476d11f80e1d763ecb723cf87165d8

Format: NVFP4 target weights.

GPUs: NVFP4 needs a Blackwell GPU (RTX 50xx / PRO).

The TP2 packages split the model across two identical GPUs; load them with tensor-parallel size 2. TP4 packages will be added later.

Check each download with sha256sum muse_glimmer_30b_nvfp4_df2q4.ginfer, sha256sum muse_glimmer_30b_nvfp4_df2q4_tp2.ginfer, sha256sum muse_glimmer_30b_nvfp4_df2.ginfer or sha256sum muse_glimmer_30b_nvfp4_df2_tp2.ginfer.

Upstream models and licenses

  • Base model: meta-models/Muse-Glimmer-30B, revision a4e59da52a7bc87ae7251dd5545c0dd437c44b68. License field apache-2.0; the repo LICENSE file is the Apache License 2.0 (full text in LICENSE here).
  • Drafter: z-lab/Muse-Glimmer-30B-DFlash2, revision b54ffdd11fa9cfe2af370012e5763d492c904128. License field apache-2.0.

Sources by part (both NVFP4 packages in this repo): The text-layer weights are RedHatAI/Muse-Glimmer-30B-NVFP4@d5109a1d (NVFP4 by Red Hat with LLM Compressor), converted to the GInfer layout without re-quantization. The Vision tower is the official BF16 from meta-models/Muse-Glimmer-30B@a4e59da5. The DFlash drafter is z-lab/Muse-Glimmer-30B-DFlash2@b54ffdd1, quantized by Sectile Labs (to Q4 in the Q4 DFlash2 packages, to NVFP4 in the NVFP4 DFlash2 packages).

Which license covers what

  • Everything derived from meta-models/Muse-Glimmer-30B and the z-lab drafter: Apache License 2.0 (LICENSE).
  • RedHatAI/Muse-Glimmer-30B-NVFP4: the revision used (d5109a1d) declared no license; the repo's current revision (e83ead54) declares Apache-2.0, the same as the Apache-2.0 base model. These weights are distributed under Apache-2.0 on that basis.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including SectileLabs/muse_glimmer_30b_nvfp4