Instructions to use xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MiMo-V2.6-Distill-Qwen-9B — MLX oQ5e-MTP

Model Details

Model Description

oQ5e-quantized MLX build with grafted Multi-Token Prediction (created by xunlinkx with the omlx pipeline) of XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B for first-class, all-local serving on Apple Silicon via omlx (the MLX OpenAI-compatible server) and mlx-lm.

Attribution chain:

Base model: XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B (official base, MIT)
MTP Donor: mlx-works/MiMo-V2.6-Distill-Qwen-9B-oQ4e-mtp (MTP head donor)
Quantization: xunlinkx's oQ5e-mtp build (this repo).

This is not a new model — it is an enhanced quantization of the existing checkpoint, produced with omlx's quantize_oq_streaming(..., oq_level=5, enhanced=True) pipeline with strict 128×512-block imatrix calibration (249 tensors applied, 0 missing, 0 mismatched) and group size 64.

The official MiMo checkpoint declares an MTP layer but does not ship MTP weights. This repository grafts the compatible native MTP head from mlx-works/MiMo-V2.6-Distill-Qwen-9B-oQ4e-mtp after strict architecture, geometry, and tokenizer-identity checks, enabling ~1.5x–2.0x speculative decoding on Apple Silicon.

Developed by: xunlinkx (oQ5e quantization + MLX build, using omlx)
Base model by: Xiaomi MiMo Team (MiMo-V2.6-Distill-Qwen-9B)
Shared by: xunlinkx
Model type: qwen3_5 hybrid SSM / linear attention text decoder + 1-layer MTP (model_type: qwen3_5)
Context length: 262,144 tokens (256K native)
License: MIT (inherits the base checkpoint license)

Compatibility

model_type: qwen3_5 loads in modern omlx and mlx-lm with native Qwen3.5 hybrid Gated DeltaNet attention and Lightning MTP support.

Validation

Validated locally before upload:

Strict model loading through the oMLX-patched MLX-LM runtime (mtp_forward operational)
Global 5-bit affine quantization metadata with group size 64
Complete strict oQe importance-matrix coverage (249 applied, 0 missing, 0 mismatched)
Tokenizer SHA-256 identity verified across source, donor, and output (06b95093...)
Official chat template verified with OpenAI function call schemas
Deterministic math smoke test (15% of 240 -> 36) generated at 33.7 tok/s (peak memory 6.40 GB)

The machine-readable validation report is included as validation.json.

Sampling (official, per the Xiaomi MiMo model card)

temperature 0.6, top_p 0.95, top_k 20, repetition_penalty 1.0
Explicit <think>...</think> chain-of-thought reasoning by default.

Usage

Using mlx-lm

mlx_lm.generate --model xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp \
  -p "Analyze the time complexity of quicksort" -m 2048 \
  --temp 0.6 --top-p 0.95

Using omlx (OpenAI-compatible server)

omlx serve --host 127.0.0.1 --port 8005
# model id: MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp

License

MIT, inherited from XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B.

Downloads last month: 90

Safetensors

Model size

10B params

Tensor type

U32

BF16

MLX

Hardware compatibility

5-bit

Model tree for xunlinkx/MiMo-V2.6-Distill-Qwen-9B-oQ5e-mtp

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Finetuned

XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B

Quantized

(68)

this model