Instructions to use m-i/Qwen3.5-397B-A17B-2.592bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use m-i/Qwen3.5-397B-A17B-2.592bit with MLX:

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

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("m-i/Qwen3.5-397B-A17B-2.592bit")
config = load_config("m-i/Qwen3.5-397B-A17B-2.592bit")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use m-i/Qwen3.5-397B-A17B-2.592bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "m-i/Qwen3.5-397B-A17B-2.592bit"

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": "m-i/Qwen3.5-397B-A17B-2.592bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use m-i/Qwen3.5-397B-A17B-2.592bit 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 "m-i/Qwen3.5-397B-A17B-2.592bit"

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 m-i/Qwen3.5-397B-A17B-2.592bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use m-i/Qwen3.5-397B-A17B-2.592bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "m-i/Qwen3.5-397B-A17B-2.592bit"

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 "m-i/Qwen3.5-397B-A17B-2.592bit" \
  --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"

Benchmarks (speed / performance)

by x-polyglot-x - opened Apr 20

Discussion

x-polyglot-x

Apr 20

Hi!

This seems interesting. I am wondering what final memory usage is like on a 128gb Mac (which is what I use). What kind of speeds do you obtain?

Thanks - it seems interesting.

m-i

Owner Apr 23

There has been so many new releases I can barely keep up. By the time I finished this upload, 3.6 moe was out. And now it seems 3.6 27b is on many aspects on par or even better than 3.5 397B, so i'll be testing this the next few days.

From the small use I had, all I can remember is that it was faster than my reading speed. But I think I'd rather use q4 with https://github.com/Anemll/anemll-flash-mlx if 3.6 27B is not workig for me.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment