Ornith-1.5-35B-A3B-MLX-3bit
MLX 3bit (affine, group size 32) quantized variant of ornith-ai/Ornith-1.5-35B-A3B (text tower quantized; vision tower and projector retained in BF16) for Apple silicon via mlx-lm.
Provenance
- Source: ornith-ai/Ornith-1.5-35B-A3B @ revision
e4dfb35a93d4b6822a811a7676f3488514abe7e2(MIT (upstreamLICENSE)). - Quantized with
mlx_lm.convert(mlx-lm 0.31.3): affine, 3-bit, group size 32.
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
pip install mlx-lm
mlx_lm.generate --model majentik/Ornith-1.5-35B-A3B-MLX-3bit --prompt "Hello"
Evaluation
| Benchmark | Score |
|---|---|
| arc_easy_acc | 0.5300 |
| hellaswag_acc | 0.4900 |
First-party tiers
ornith-ai publishes first-party 4bit/6bit/8bit MLX and GGUF packs for this model. The tiers in this family (2bit, 3bit, 5bit, MXFP4) complement those first-party releases rather than duplicating them.