Brooooooklyn/Qwen3.5-35B-A3B-UD-Q2_K_XL-mlx

Qwen3.5-35B-A3B — UD-Q2KXL (mlx-node)

2-bit base mixed-precision quantization of Qwen/Qwen3.5-35B-A3B for Apple Silicon, using the **Unsloth Dynamic** quantization strategy via mlx-node.

All Variants

Benchmarked on Apple M3 Max 128GB, multi-turn chat (Turn 4 decode, steady-state).

Performance

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).

Per-Tensor Bit Assignments (N=2)

Quantization Strategy

Based on Unsloth Dynamic 2.0 per-tensor KLD analysis. Sensitive layers get higher bits with AWQ correction, while the bulk of FFN expert weights are aggressively quantized. imatrix AWQ pre-scaling amplifies important weight channels and fuses inverse scales into preceding layer norms (zero inference overhead).

AWQ-correctable projections (q/k/v, inprojqkv/z) are quantized at 4-bit via input_layernorm. Non-AWQ-correctable projections (oproj, outproj) are kept at bf16 — their inputs come from attention/GDN computation, not from a norm layer.

Architecture

Usage

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

const model = await loadModel('./Qwen3.5-35B-A3B-UD-Q2_K_XL-mlx');

const result = await model.chat(
  [{ role: 'user', content: 'Explain the hybrid attention mechanism in Qwen3.5.' }],
  { maxNewTokens: 2048, temperature: 0.6, enableThinking: false },
);
console.log(result.text);

How It Was Made

mlx convert \
  -i Qwen3.5-35B-A3B \
  -o Qwen3.5-35B-A3B-UD-Q2_K_XL-mlx \
  -q --q-bits 2 --q-recipe unsloth \
  --imatrix-path imatrix_unsloth.gguf

Acknowledgments

• —[Unsloth](https://unsloth.ai) — Quantization strategy based on their per-layer KLD benchmarks and Dynamic 2.0 methodology
• —[Qwen Team](https://huggingface.co/Qwen) — For the Qwen3.5 model family
• —[Apple MLX](https://github.com/ml-explore/mlx) — For the Metal-accelerated ML framework

License

Apache 2.0 (inherited from base model).