Brooooooklyn/Ornith-1.0-35B-UD-Q8_K_XL-mlx

Ornith-1.0-35B — UD-Q8KXL (mlx-node)

8-bit base mixed-precision quantization of deepreinforce-ai/Ornith-1.0-35B for Apple Silicon, using the Unsloth Dynamic per-tensor bit allocation (without imatrix AWQ) via mlx-node.

Ornith-1.0 is a self-improving family of open-source agentic coding models. The 35B member is a Qwen3.5-VL-MoE (hybrid Gated-DeltaNet + full attention, 256 experts, vision-language) post-train.

All Variants

Benchmarked on a cool Apple M5 Max: median decode throughput over three 512-token generations, with a 60-second idle GPU cooldown after every generation. (Sustained decode on Apple Silicon is thermally sensitive — back-to-back benchmarking on a hot chip can understate throughput by 20–30%, so every model here was measured from a comparable cool start.)

Performance

Steady-state decode: 91.5 tok/s (1.5x vs BF16) on Apple M5 Max. Decode is memory-bandwidth bound on Apple Silicon — fewer bytes per token directly translates to higher throughput. The MoE architecture activates only 8 of 256 experts per token (~3B active out of 35.9B total), so the active-weight footprint streamed per token is what matters.

Output Quality

Decoded-text quality was verified against the BF16 reference with a multi-judge review of the actual generated output (not a heuristic): a 4-turn factual chat plus a Python is_balanced() bracket-matching task. This UD-Q8_K_XL build produced coherent prose, correct facts, and a correct implementation — no runaway generation, repetition loops, or stray tokens — on par with full precision. (The 2-bit tier is intentionally excluded from this collection: it was the only width that showed coherence breakdown.)

Per-Tensor Quantization

Quantization Strategy

Built on Unsloth Dynamic 2.0 per-tensor KLD analysis: sensitive layers (attention/SSM inputs, downproj, embeddings/head) get higher bits, while the bulk of FFN expert weights are quantized to the base width. `selfattn.oproj`, `linearattn.outproj`, the split low-rank GDN projections (`inproja/b`) and the MoE router gates are pinned to 8-bit affine (groupsize 64). GatedDeltaNet state-space parameters and the vision encoder stay bf16.

Note: These ornith quants apply the Unsloth bit allocation without imatrix AWQ pre-scaling — ornith has no published imatrix, so the attention/SSM channels are quantized directly. Expect a small quality gap versus an imatrix-calibrated build at the lowest bit widths.

Architecture

Usage

import { loadSession } from '@mlx-node/lm';

const session = await loadSession('./Ornith-1.0-35B-UD-Q8_K_XL-mlx');

for await (const event of session.sendStream('Write a Python function to merge two sorted lists.', {
  config: { maxNewTokens: 2048, temperature: 0.6, reasoningEffort: 'low' },
})) {
  if (!event.done) process.stdout.write(event.text);
}

How It Was Made

mlx convert \
  -i Ornith-1.0-35B \
  -o Ornith-1.0-35B-UD-Q8_K_XL-mlx \
  -q --q-recipe unsloth --q-bits 8

The Unsloth recipe's per-tensor bit tiers were applied without imatrix AWQ (no native ornith imatrix). 7-bit tiers are snapped up to 8-bit (MLX affine supports 2/3/4/5/6/8-bit).

Acknowledgments

• —[Unsloth](https://unsloth.ai) — Per-layer KLD bit-allocation strategy (Dynamic 2.0)
• —[DeepReinforce](https://deep-reinforce.com/ornith.html) — For the Ornith-1.0 model family
• —[Qwen Team](https://huggingface.co/Qwen) — For the Qwen3.5 base architecture
• —[Apple MLX](https://github.com/ml-explore/mlx) — For the Metal-accelerated ML framework

License

MIT (inherited from base model).