Instructions to use blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit 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("blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit")
config = load_config("blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit")

# 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 blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit"

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": "blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit 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 "blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit"

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 blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit"

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 "blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit" \
  --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"

Qwen3.8-27B-TURBO-Heretic-NM-DAU (MLX 8-bit)

MLX conversion of DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU (a heretic finetune of Qwen/Qwen3.8-27B). The language model was converted text-only with mlx-lm, the source checkpoint's BF16 vision tower was grafted back in as vision_tower.*, the result validated for the Splash engine (family Qwen3.8-27B), and the chat template swapped to Sharp v22.5.0.

Quantization

Language model linears (incl. lm_head): affine 8-bit, group size 64
Token embeddings, GDN A_log/dt_bias, norms: BF16
Vision tower (vision_tower.*, 333 tensors): BF16, unquantized — grafted from the source checkpoint's model.visual.* weights
MTP weights from the source checkpoint were dropped

Bits per weight: 8.501. Total size: ~28 GB.

Layout

7 safetensors shards: shard 1 holds the vision tower, shards 2–7 the language model, matching the layout of mlx-community/Qwen3.8-27B-4bit.

Build

mlx_lm convert -q --q-bits 8 (text-only output; mlx-lm 0.32.0 drops the model.visual tower for qwen3_5)
Tower grafted byte-for-byte from the source BF16 shards as vision_tower.*, vision_config and processor files restored from the source repo

Chat template

Sharp v22.5.0 (qwen3.8-froggeric-v22.5.0) replaces the stock Qwen VL template: terseness block appended after your system prompt, thinking retention for prefix-cache hits, tool-call token parity. Per-request knobs via chat_template_kwargs: {"terse": false}, reasoning_effort, enable_thinking.

Splash

Passes the Splash engine model-check as family Qwen3.8-27B:

splash serve --model blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit

Downloads last month: 112

Safetensors

Model size

27B params

Tensor type

U32

BF16

MLX

Hardware compatibility

8-bit

Model tree for blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit

Base model

Qwen/Qwen3.8-27B

Finetuned

DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU

Quantized

(102)

this model