Qwen3.5-397B-A17B-TurboQuant-MLX-6bit

6-bit MLX weight-quantized build of Qwen/Qwen3.5-397B-A17B (397B total / 17B active Sparse MoE, multimodal) prepared with TurboQuant randomized Hadamard rotations. Optimized for Apple Silicon via MLX.

6-bit hits a sweet spot between 8-bit quality and 4-bit compactness — ideal when you want headroom for long context on a 512 GB Mac without the last percentage points of fidelity loss seen at 4-bit.

Quickstart

from mlx_lm import load, generate

model, tokenizer = load("majentik/Qwen3.5-397B-A17B-TurboQuant-MLX-6bit")

prompt = tokenizer.apply_chat_template(
    [{"role": "user", "content": "Summarize the plot of Dune in 3 sentences."}],
    add_generation_prompt=True,
)
print(generate(model, tokenizer, prompt=prompt, max_tokens=256, verbose=True))

Model Specs

Property	Value
Base model	Qwen/Qwen3.5-397B-A17B
Architecture	Sparse Mixture-of-Experts (MoE)
Total parameters	397B
Active per token	17B
Modalities	Image + Text → Text (image-text-to-text)
Context window	256K tokens
Weight quantization	6-bit MLX (TurboQuant pre-rotation)
Approx. disk footprint	~300 GB
License	Apache 2.0

RotorQuant vs TurboQuant

Aspect	TurboQuant (this repo)	RotorQuant
Rotation	Randomized Hadamard (static)	Learned orthogonal rotors (data-calibrated)
Calibration	Zero-shot	~512 sample calibration pass
Accuracy @ 6-bit	~99.6% of FP16 baseline	~99.8% of FP16 baseline
Best for	Fastest turnaround	Highest fidelity at same bit-width

Memory Estimates (6-bit MLX)

Context	Active memory (approx.)
8K	~308 GB
32K	~318 GB
128K	~348 GB
256K	~378 GB

Hardware Requirements

• Minimum: Apple Silicon with 384 GB+ unified memory (tight)
• Recommended: 512 GB unified memory Mac Studio
• Does not fit on 96 GB / 128 GB / 192 GB / 256 GB Macs

See Also

• TurboQuant MLX: 8-bit · 5-bit · 4-bit · 2-bit
• RotorQuant MLX 6-bit: majentik/Qwen3.5-397B-A17B-RotorQuant-MLX-6bit
• Base model: Qwen/Qwen3.5-397B-A17B