SectileLabs/muse_glimmer_30b_smol

Muse-Glimmer-30B, GInfer SMOL 2-bit mix

muse_glimmer_30b_smol_df2.ginfer is a single-file GInfer package of Muse-Glimmer-30B for the GInfer inference engine (and GChat, which is built on it). It runs only on GInfer / GChat. It is not a safetensors or GGUF file and other runtimes cannot load it.

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

Package

File	muse_glimmer_30b_smol_df2.ginfer
Size	13,838,508,544 bytes
SHA-256	9d4ca1b204b993e3a5dd9e52c8eda0d804801dd4ded664dd37b81641fcb44a2c
Format	SMOL 2-bit mix (calibrated map: Q4 attention and late MLP-down layers, W8 key/value, Q2 elsewhere); Q4 DFlash2 drafter included; tensor-parallel size 1 (TP1)
GPUs	SMOL runs on NVIDIA Ampere, Ada and Blackwell GPUs.

Check the download with sha256sum muse_glimmer_30b_smol_df2.ginfer. TP2/TP4 packages of the same format will be added to this repo later.

Known behaviour

This package may repeat itself in long generations at higher context lengths. For long creative text, use the INT or NVFP4 package of Muse-Glimmer-30B where it fits your GPU.

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.

Quantized by Sectile Labs to SMOL 2-bit mix from the official BF16 weights of meta-models/Muse-Glimmer-30B (revision a4e59da52a7bc87ae7251dd5545c0dd437c44b68) using a calibrated sensitivity map (AutoRound 0.15.0), then converted to a .ginfer package.

This package is a conversion of the upstream weights into the GInfer format; GInfer does not change the upstream license. The upstream license text is in LICENSE.