Instructions to use aquaduck/Qwen3.8-27B-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use aquaduck/Qwen3.8-27B-MLX 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("aquaduck/Qwen3.8-27B-MLX")

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 aquaduck/Qwen3.8-27B-MLX with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "aquaduck/Qwen3.8-27B-MLX"

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": "aquaduck/Qwen3.8-27B-MLX"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use aquaduck/Qwen3.8-27B-MLX with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "aquaduck/Qwen3.8-27B-MLX"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "aquaduck/Qwen3.8-27B-MLX"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "aquaduck/Qwen3.8-27B-MLX",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use aquaduck/Qwen3.8-27B-MLX 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 "aquaduck/Qwen3.8-27B-MLX"

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 aquaduck/Qwen3.8-27B-MLX

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use aquaduck/Qwen3.8-27B-MLX with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "aquaduck/Qwen3.8-27B-MLX"

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 "aquaduck/Qwen3.8-27B-MLX" \
  --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"

Model Card for aquaduck/Qwen3.8-27B-MLX

Pinned 4bit MLX of Qwen3.8-27B (qwen/qwen3.8-27b), plus midpoint layer shards for staged / multi-node loading (Aquaduck Arc mlx-package-v1). The shard files are not a new quantization. They are contiguous midpoint packages cut from the full 4bit MLX in this repo.

Model lineage

Qwen/Qwen3.8-27B └── quantized → Qwen/Qwen3.8-27B (4bit) └── full MLX + midpoint shards → aquaduck/Qwen3.8-27B-MLX (this repo)

Base weights: https://huggingface.co/Qwen/Qwen3.8-27B (apache-2.0)
Quantization source: https://huggingface.co/Qwen/Qwen3.8-27B (tag 4bit)
This repo: full 4bit MLX (model-00001-of-00003.safetensors) and midpoint MLX shards

Model Details


Catalog id	`qwen/qwen3.8-27b`
Quantization	4bit
Parameters	27.8B
Native context	262,144 tokens
License	apache-2.0
Base model	Qwen/Qwen3.8-27B
Ingest MLX	Qwen/Qwen3.8-27B

Model Description

Hosted by: Aquaduck (hosting and layer packaging only; base model by Qwen Team / Alibaba Cloud; MLX quant by mlx-lm)
Shared by: Aquaduck AI
Model type: Causal language model (Qwen3.8-27B), MLX 4bit
Language(s): Multilingual (same as base)
License: apache-2.0 (inherits from Qwen/Qwen3.8-27B)
Finetuned from model: N/A — not a fine-tune
Derived from: Qwen/Qwen3.8-27B ← Qwen/Qwen3.8-27B

Model Sources

Base model card: https://huggingface.co/Qwen/Qwen3.8-27B
Quantized MLX source: https://huggingface.co/Qwen/Qwen3.8-27B

Files

File	Role	Approx. size
`model-00001-of-00003.safetensors`	Full-model MLX (4bit)	~15.13 GB
`layers-0-32/model.safetensors`	Split shard (layers 0–31)	~7.57 GB
`layers-32-64/model.safetensors`	Split shard (layers 32–63)	~8.28 GB

Total layers: 64
Valid split boundaries: 32 Filenames use exclusive end indices (layers-{start}-{endExclusive}/model.safetensors).

Uses

Direct Use

Full model-00001-of-00003.safetensors: standard single-file 4bit MLX (mlx-lm-compatible). Use this for single-node / local runs.
layers-*-*/model.safetensors: Aquaduck / Arc staged loading only. These are not drop-in complete models for stock mlx-lm. Use the base model’s chat template (including thinking / instruct modes as documented on the base model card); other formats will not work correctly.

Out-of-Scope Use

Expecting any one shard to run as a complete model
Treating this repo as a new training run or re-quant
Uses prohibited by the apache-2.0 license or the base model’s model card guidance

Bias, Risks, and Limitations

Same capabilities, biases, and risks as Qwen/Qwen3.8-27B. 4bit quantization can degrade quality vs. the original higher-precision releases. Layer sharding does not change weights beyond packaging.

Recommendations

Follow the base model’s docs for chat template, thinking vs instruct modes, and sampling. Prefer model-00001-of-00003.safetensors in this repo when you do not need staged loading.

How to Get Started

These files are meant to be loaded automatically by the Aquaduck desktop app.

Download the Aquaduck desktop app and sign in.
Devices connected to the internet will receive a model assignment from the model catalog (qwen/qwen3.8-27b).
Download the model from the Home view. The app will:
- download only the assigned file from this repo (full model-00001-of-00003.safetensors or one midpoint half)
- keep that stage ready for serving You do not need to pick files by hand, but you may for local serving. Assignment and download are driven by model catalog metadata. The full model-00001-of-00003.safetensors is a standard 4bit MLX. The layers-*-*/model.safetensors files are not.

Training Details

No training. Weights come from Qwen Team / Alibaba Cloud; 4bit MLX from Qwen/Qwen3.8-27B; this repo hosts that MLX and (when split) packages it into midpoint layer shards.

Evaluation

No separate evals for the hosted MLX or shards. See Qwen/Qwen3.8-27B.

Technical Specifications

Architecture: Qwen3.8-27B (~27.8B params, GQA (24 Q / 4 KV heads), 64 layers, hidden dim 5120)
Quantization: 4bit
Packaging: pinned full 4bit MLX; optional Arc midpoint shards (layers-{start}-{endExclusive}/model.safetensors)
Package format: mlx-package-v1
Split: 2 stages at layer 32 (maxStages: 2)

Citation

@misc{qwen3827b,
    title  = {Qwen3.8-27B},
    author = {Qwen Team / Alibaba Cloud},
    year   = {2026},
    url    = {https://huggingface.co/Qwen/Qwen3.8-27B}
}

Credit: