Qwen3.6-27B — UD-Q6_K_XL (mlx-node)

6-bit base mixed-precision quantization of Qwen/Qwen3.6-27B for Apple Silicon, using the Unsloth Dynamic quantization strategy via mlx-node.

	Original (BF16)	This Model
Size	~51 GB	27 GB
Format	SafeTensors (sharded)	SafeTensors (sharded)
Precision	BF16 uniform	Mixed 6-bit + BF16

All Variants

Repo	GGUF Equivalent	Size	Decode (tok/s)	Speedup vs BF16
Brooooooklyn/Qwen3.6-27B-UD-Q2_K_XL-mlx	UD-Q2_K_XL	15 GB	18.6	3.32x
Brooooooklyn/Qwen3.6-27B-UD-Q3_K_XL-mlx	UD-Q3_K_XL	18 GB	15.5	2.77x
Brooooooklyn/Qwen3.6-27B-UD-Q4_K_XL-mlx	UD-Q4_K_XL	21 GB	13.9	2.48x
Brooooooklyn/Qwen3.6-27B-UD-Q5_K_XL-mlx	UD-Q5_K_XL	25 GB	12.0	2.14x
Brooooooklyn/Qwen3.6-27B-UD-Q6_K_XL-mlx	UD-Q6_K_XL	27 GB	10.8	1.93x
Brooooooklyn/Qwen3.6-27B-UD-Q8_K_XL-mlx	UD-Q8_K_XL	30 GB	9.9	1.77x

Benchmarked on Apple M3 Max 128GB via examples/lm.ts (Turn 4 steady-state decode).

Performance

Model	Size	Decode (tok/s)	Speedup
BF16 (unquantized)	51 GB	5.6	baseline
This model (UD-Q6_K_XL)	27 GB	10.8	1.93x faster

Decode is memory-bandwidth bound on Apple Silicon — fewer bytes per token directly translates to higher throughput. The hybrid architecture interleaves linear attention (gated delta net, 48/64 layers) with full attention (16/64 layers).

Per-Tensor Bit Assignments (N=6)

Weight	Bits	Rationale
`embed_tokens`	8-bit	KLD ~0.15 — very low sensitivity
`lm_head`	8-bit	KLD ~0.05 — safest tensor
`self_attn.q/k/v_proj`	8-bit + AWQ	KLD ~1.5–2.9, AWQ via layernorm
`linear_attn.in_proj_qkv/z`	8-bit + AWQ	KLD ~2.9, AWQ via layernorm
`self_attn.o_proj`	bf16	NOT AWQ-correctable
`linear_attn.out_proj`	bf16	KLD ~6.0 — worst tensor
`down_proj`	8-bit	"Slightly more sensitive" (snap N+1=7 → 8)
`gate_proj`, `up_proj`	6-bit	base bits
GDN params (A_log, etc)	bf16	State-space dynamics

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 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, in_proj_qkv/z) are quantized at 8-bit via input_layernorm. Non-AWQ-correctable projections (o_proj, out_proj) are kept at bf16 — their inputs come from attention/GDN computation, not from a norm layer.

Architecture

Parameter	Value
Total parameters	27.4B (dense — all active)
Hidden size	5,120
Layers	64 (48 linear + 16 full attention)
Attention heads	24 (4 KV heads, GQA 6:1)
Head dimension	256
Intermediate size	17,408
Vocab size	248,320
Max context	262,144 tokens

Usage

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

const session = await loadSession('./Qwen3.6-27B-UD-Q6_K_XL-mlx');

for await (const event of session.sendStream('Explain the hybrid attention mechanism in Qwen3.6.', {
  config: { maxNewTokens: 2048, temperature: 0.6, reasoningEffort: 'low' },
})) {
  if (!event.done) process.stdout.write(event.text);
}

How It Was Made

mlx convert \
  -i Qwen3.6-27B \
  -o Qwen3.6-27B-UD-Q6_K_XL-mlx \
  -q --q-bits 6 --q-recipe unsloth \
  --imatrix-path imatrix_unsloth.gguf

Acknowledgments

Unsloth — Quantization strategy based on their per-layer KLD benchmarks and Dynamic 2.0 methodology
Qwen Team — For the Qwen3.6 model family
Apple MLX — For the Metal-accelerated ML framework

License

Apache 2.0 (inherited from base model).

Brooooooklyn/Qwen3.6-27B-UD-Q6_K_XL-mlx