Instructions to use Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit 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("Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit")

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 Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit 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 "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"

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 Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit

Run Hermes

hermes

OpenClaw new

How to use Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"

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 "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit" \
  --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"

MLX LM

How to use Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Qwen3-4B-Computer-Science-MLX-4bit

Overview

Qwen3-4B-Computer-Science-MLX-4bit is a 4-bit MLX conversion of Qwen3-4B-Computer-Science for inference on Apple Silicon.

The model was converted from the original BF16 Safetensors release using mlx-lm. It is intended for local inference on supported Mac systems using the MLX framework.

Model Information

Property	Value
Source model	Irfanuruchi/Qwen3-4B-Computer-Science
Base architecture	Qwen3-4B
Framework	MLX
Quantization	4-bit
Quantization group size	64
Effective bits per weight	4.501
Weight format	Safetensors
Primary platform	Apple Silicon
Language	English
License	Apache-2.0

Training Data

The source model was instruction-tuned using permissively licensed datasets.

Dataset	Configuration	License
HuggingFaceTB/smoltalk	smol-magpie-ultra	Apache-2.0
agentica-org/DeepCoder-Preview-Dataset	primeintellect	MIT

Dataset Size

Split	Samples
Training	60,989
Evaluation	512

Intended Use

This model is intended for:

Software engineering
Programming assistance
Code generation
Debugging
Code review
Algorithm implementation
Computer science education
General technical reasoning

Installation

Install MLX-LM:

python -m pip install mlx-lm

Usage

Command Line

mlx_lm.generate \
  --model Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit \
  --prompt "Write a Python function that returns the first n Fibonacci numbers." \
  --max-tokens 200

Python

from mlx_lm import generate, load

model, tokenizer = load(
    "Irfanuruchi/Qwen3-4B-Computer-Science-MLX-4bit"
)

response = generate(
    model,
    tokenizer,
    prompt="Write a Python function that returns the first n Fibonacci numbers.",
    max_tokens=200,
)

print(response)

Conversion

The model was converted from:

Irfanuruchi/Qwen3-4B-Computer-Science

Conversion configuration:

Parameter	Value
Quantization enabled	Yes
Quantization bits	4
Quantization group size	64
Effective bits per weight	4.501

Release Validation

The release was validated locally on Apple Silicon.

Test	Result
MLX conversion	Passed
Model loading	Passed
Text generation	Passed
Generation speed	53.512 tokens/sec
Peak memory	2.356 GB

Performance measurements are from one local generation test and may vary by device, prompt, context length, and software version.

Release Artifacts

model.safetensors
model.safetensors.index.json
config.json
generation_config.json
tokenizer.json
tokenizer_config.json
added_tokens.json
special_tokens_map.json
merges.txt
vocab.json
README.conversion.md
SHA256SUMS
LICENSE
README.md

Integrity Verification

Verify the downloaded files on macOS:

shasum -a 256 -c SHA256SUMS

On Linux:

sha256sum -c SHA256SUMS

Limitations

Quantization can affect output quality relative to the BF16 checkpoint.
The model may produce incorrect or incomplete code.
Generated code should be reviewed and tested before use.
Performance depends on the Apple Silicon device and available memory.
This release is intended for MLX-compatible systems.

License

The model is distributed under the Apache License 2.0.

The source model is based on Qwen3-4B, which is also distributed under the Apache License 2.0.

Acknowledgements

Alibaba Qwen Team
Apple MLX Team
Hugging Face
SmolTalk contributors
DeepCoder contributors

Citation

@software{uruci2026qwen3computersciencemlx,
  title={Qwen3-4B-Computer-Science-MLX-4bit},
  author={Irfan Uruçi},
  year={2026},
  publisher={Hugging Face}
}