CoolFace
Modelpublic

majentik/Qwen3.8-27B-MLX-6bit

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face
1likes1.2kdownloads
Model Card

Qwen3.8-27B-MLX-6bit

MLX 6bit (affine, group size 64) quantized variant of Qwen/Qwen3.8-27B (text tower quantized; vision tower and projector retained in BF16) for Apple silicon via mlx-lm.

Provenance

  • —Source: Qwen/Qwen3.8-27B @ revision 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0 (Apache-2.0 (upstream LICENSE)).
  • —Quantized with mlx_lm.convert (mlx-lm 0.31.3): affine, 6-bit, group size 64.

Smoke gate

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.

Usage

bash
pip install mlx-lm
mlx_lm.generate --model majentik/Qwen3.8-27B-MLX-6bit --prompt "Hello"

Evaluation

BenchmarkScore
arceasyacc0.6000
hellaswag_acc0.5400

License

Apache-2.0 — see the upstream LICENSE file in Qwen/Qwen3.8-27B.

Available tiers

RepoQuant
majentik/Qwen3.8-27B-MLX-2bit2bit
majentik/Qwen3.8-27B-MLX-3bit3bit
majentik/Qwen3.8-27B-MLX-4bit4bit
majentik/Qwen3.8-27B-MLX-5bit5bit
majentik/Qwen3.8-27B-MLX-6bit6bit
majentik/Qwen3.8-27B-MLX-8bit8bit
majentik/Qwen3.8-27B-MLX-MXFP4MXFP4