Instructions to use KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K 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("KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K")
config = load_config("KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K")

# 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
Unsloth Desktop

How to use KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K"

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": "KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K 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 "KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K"

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 KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K"

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 "KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K" \
  --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"

Qwen3.6-35B-A3B-MLX-UD-8bit-512K

Extended context version of unsloth/Qwen3.6-35B-A3B-MLX-8bit with 512K context window (up from 256K).

What Changed

Context window extended from 256K → 512K using YaRN RoPE scaling (factor 2.0x).

Only config.json was modified — model weights are identical to the original.

Config Changes

"max_position_embeddings": 524288,
"rope_parameters": {
    "mrope_interleaved": true,
    "mrope_section": [11, 11, 10],
    "partial_rotary_factor": 0.25,
    "rope_theta": 10000000,
    "rope_type": "yarn",
    "factor": 2.0,
    "original_max_position_embeddings": 262144,
    "beta_fast": 32,
    "beta_slow": 1,
    "mscale": 1,
    "mscale_all_dim": 0
}

Why This Works Well

This model has a hybrid architecture that is naturally suited for long context:

40 layers total: 30 linear attention (no position limit) + 10 full attention (RoPE)
Only 25% of layers need RoPE scaling
partial_rotary_factor: 0.25 — only 64 out of 256 head dimensions use rotary embeddings
Already high rope_theta: 10,000,000

Usage

# Install dependencies
curl -fsSL https://raw.githubusercontent.com/unslothai/unsloth/refs/heads/main/scripts/install_qwen3_6_mlx.sh | sh
source ~/.unsloth/unsloth_qwen3_6_mlx/bin/activate

# Run with 512K context
python -m mlx_vlm.chat --model KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K

Memory Requirements

Model weights: ~20GB (8-bit)
KV cache at 512K: ~5-6GB (only 10 full attention layers with GQA)
Total: ~26-30GB unified memory recommended

Credits

Original model: Qwen/Qwen3.6-35B-A3B
MLX quantization: unsloth/Qwen3.6-35B-A3B-MLX-8bit
Context extension: YaRN RoPE scaling applied by KCh3dRi4n

Downloads last month: 73

Safetensors

Model size

10B params

Tensor type

BF16

U32

MLX

Hardware compatibility

8-bit

Model tree for KCh3dRi4n/Qwen3.6-35B-A3B-MLX-UD-8bit-512K

Base model

Qwen/Qwen3.6-35B-A3B

Quantized

(776)

this model