mlx-community/MiMo-V2.6-Distill-Qwen-9B-OptiQ-4bit

Built with mlx-optiq, the MLX-native toolkit to quantize, fine-tune, and serve LLMs locally on Apple Silicon, no PyTorch and no cloud. Try the Lab · All OptiQ quants · Docs

A 4-bit mixed-precision MLX quant of XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B, a distillation of the Qwen3.5 9B architecture. Sensitive layers are kept at 8-bit and robust ones at 4-bit, rather than crushing everything to a uniform width.

Quantization details

Property	Value
Predominant precision	4-bit
Layers at 8-bit	134
Layers at 4-bit	116
Bits per weight	6.34
Size on disk	6.77 GB (from ~18 GB bf16)
Group size	64

How the bit-widths were chosen

Honestly stated, because it differs from most OptiQ quants: the per-layer allocation was not measured on this model. It was transferred from mlx-community/Qwen3.5-9B-OptiQ-4bit, whose allocation came from a KL-divergence sensitivity pass over a six-domain calibration mix (prose, reasoning, code, agent, tool-call, instructions).

That transfer is sound here because the two models share an architecture exactly — qwen3_5, 32 layers with 8 full-attention and 24 linear-attention, 16 heads, 4 KV heads, head_dim 256, hidden 4096, vocab 248,320 — so every layer in the recipe has a counterpart with the same role and shape. All 250 tensors matched with none unmatched, which is the check that matters: an unmatched tensor would silently fall back to flat 4-bit and make this a uniform quant wearing a mixed-precision name.

What sensitivity measures is how much a layer's role in the architecture suffers from precision loss. What it cannot know is whether this model's own training moved that sensitivity around. If you are quantizing your own fine-tune and want the allocation measured against it, run optiq convert and let the sweep do it.

What was verified

• 250/250 tensors matched the recipe, 0 unmatched.
• Generation compared against the bf16 parent on identical prompts — factual recall, arithmetic with working shown, an iterative Fibonacci implementation, and a technical explanation. The answers agree; the quant runs 2.4–7× faster.
• OptiQ's release contract (artifact layout, metadata, mixed-precision assertions).

Not run for this model: the six-metric Capability Score. The published scores for the Qwen3.5-9B quant describe that model, not this one, and are not claimed here.

Use it

pip install mlx-optiq
optiq serve --model mlx-community/MiMo-V2.6-Distill-Qwen-9B-OptiQ-4bit

Or with mlx-lm directly:

from mlx_lm import generate, load

model, tokenizer = load("mlx-community/MiMo-V2.6-Distill-Qwen-9B-OptiQ-4bit")
prompt = tokenizer.apply_chat_template(
    [{"role": "user", "content": "Explain write-ahead logging in two sentences."}],
    add_generation_prompt=True, tokenize=False)
print(generate(model, tokenizer, prompt=prompt, max_tokens=256))

The per-layer bit map is in optiq/metadata.json and in the quantization block of config.json.