MLX 4bit (affine, group size 32) quantized variant of meta-models/Muse-Glimmer-30B (text tower quantized; vision tower and projector retained in BF16) for Apple silicon via mlx-lm.

Before upload this pack passed a deterministic coherence gate: greedy 48-token chat generation loaded through mlx_lm.load, judged for emptiness, repetition loops, multi-script gibberish, and special-token debris. Verdict: ok.

Text-benchmark scores are not published for this pack: the muse_glimmer architecture is not loadable by stock mlx_lm, which the eval harness uses. Quality gating is the deterministic coherence smoke above (greedy 48-token generation via mlx_vlm), which this tier passed before upload.

Apache-2.0 — see the upstream LICENSE file in meta-models/Muse-Glimmer-30B.