Instructions to use ukuwzi/Qweble-Sol-4B-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ukuwzi/Qweble-Sol-4B-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("ukuwzi/Qweble-Sol-4B-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 ukuwzi/Qweble-Sol-4B-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 "ukuwzi/Qweble-Sol-4B-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": "ukuwzi/Qweble-Sol-4B-MLX"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use ukuwzi/Qweble-Sol-4B-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 "ukuwzi/Qweble-Sol-4B-MLX"

Run an OpenAI-compatible server

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

Hermes Agent

How to use ukuwzi/Qweble-Sol-4B-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 "ukuwzi/Qweble-Sol-4B-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 ukuwzi/Qweble-Sol-4B-MLX

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ukuwzi/Qweble-Sol-4B-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 "ukuwzi/Qweble-Sol-4B-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 "ukuwzi/Qweble-Sol-4B-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"

Qweble-Sol 4B — MLX

Qweble-Sol logo

Qweble-Sol is a fine-tuned derivative of Qwen3.5-4B focused on systems administration, programming, troubleshooting, DevOps, and technical assistance.

This repository contains the 4-bit MLX build of Qweble-Sol, optimized for local inference on Apple Silicon with mlx-lm.

Base model: Qwen3.5-4B
Base model developer: Alibaba / Qwen Team
Fine-tuned and packaged by: ukuwzi
Parameters: ~4.21B
Format: MLX
Quantization: 4-bit
Primary runtime: mlx-lm
Primary platform: Apple Silicon
Primary use: Systems administration, coding, debugging, networking, automation, and DevOps

Qweble-Sol is an independent fine-tune and is not an official Alibaba or Qwen model.

Quick Start

Install MLX-LM:

pip install -U mlx-lm

Run directly from Hugging Face:

mlx_lm.chat --model ukuwzi/Qweble-Sol-4B-MLX

For one-shot generation:

mlx_lm.generate \
  --model ukuwzi/Qweble-Sol-4B-MLX \
  --prompt "Explain Docker in one paragraph." \
  --max-tokens 256

Overview

Qweble-Sol is intended to be a compact local technical assistant for tasks such as:

Linux and macOS administration
Shell scripting
Python, Go, Rust, and JavaScript
Debugging and code review
Docker and container workflows
Networking and troubleshooting
Server configuration
Automation
CI/CD and DevOps
Security best practices
General technical questions

The fine-tune is intended to preserve the general reasoning and conversational abilities of Qwen3.5-4B while adapting the model toward practical systems-oriented and programming workflows.

Qweble-Sol retains Qwen3.5's reasoning/thinking behavior.

Model Lineage

Qwen3.5-4B
    │
    ▼
MLX fine-tuning
    │
    ▼
Qweble-Sol
    │
    ▼
4-bit MLX quantization
    │
    ▼
Qweble-Sol 4B — MLX 4-bit

The original Qwen3.5-4B base model was developed by the Qwen Team at Alibaba.

Qweble-Sol was fine-tuned and packaged by ukuwzi.

Why MLX 4-bit?

This build is intended for users who want a smaller and more memory-efficient version of Qweble-Sol on Apple Silicon.

Advantages include:

Lower memory usage than the full 16-bit model
Faster loading
Smaller storage footprint
Native Apple Silicon inference with MLX
No GGUF conversion required
Good fit for laptops and systems with limited unified memory

For maximum fidelity, a higher-precision version may be preferable.

For Ollama or llama.cpp, use the separate GGUF Q4_K_M release.

Usage

Interactive Chat

mlx_lm.chat --model ukuwzi/Qweble-Sol-4B-MLX

Example:

>> Explain how to troubleshoot a Docker container that keeps restarting.

Text Generation

mlx_lm.generate \
  --model ukuwzi/Qweble-Sol-4B-MLX \
  --prompt "Write a Bash script that reports disk usage above 80 percent." \
  --max-tokens 512

Suggested System Prompt

You are Qweble-Sol, a local systems administration and coding assistant.

Base model: Qwen3.5-4B by Alibaba/Qwen.
Fine-tuned and packaged by: ukuwzi.

Qweble-Sol is the fine-tuned derivative model.
Qwen3.5-4B is the base model.

You help with:
- Linux and macOS administration
- Networking and troubleshooting
- Docker and containers
- Server configuration
- Python, Go, Rust, JavaScript, and shell scripting
- Debugging and code review
- Automation and DevOps
- Security best practices

Be technically precise, practical, and concise.

Depending on your installed mlx-lm version, you can pass it with:

mlx_lm.chat \
  --model ukuwzi/Qweble-Sol-4B-MLX \
  --system-prompt "You are Qweble-Sol, a precise systems administration and coding assistant."

Performance

Performance depends heavily on:

Apple Silicon generation
Unified memory capacity
Prompt length
Context length
Sampling settings
mlx-lm version

The exact published 4-bit MLX build should be benchmarked separately from the 16-bit model and GGUF release.

Recommended Benchmark Command

mlx_lm.generate \
  --model ukuwzi/Qweble-Sol-4B-MLX \
  --prompt "Explain Docker in one paragraph." \
  --max-tokens 256

Record:

Prompt throughput
Generation throughput
Peak memory

Once measured, results can be added here.

Other Builds

Ollama / GGUF

Qweble-Sol is also available as a Q4_K_M GGUF build.

Run directly with Ollama:

ollama run ukuwzi/qweble-sol:4b-q4_k_m

Ollama page:

https://ollama.com/ukuwzi/qweble-sol

llama.cpp

The GGUF build can also be used with llama.cpp.

Example:

./llama-cli \
  -m Qweble-Sol-4B-Q4_K_M.gguf

Intelligence / Quality Benchmarks

Formal quality benchmarks are still being evaluated.

Planned evaluations include:

Benchmark	Focus	Qweble-Sol
MMLU-Pro	General knowledge + reasoning	TBD
AIME	Mathematical reasoning	TBD
MBPP / MBPP+	Python programming	TBD
LiveCodeBench	Code generation	TBD
Sysadmin Eval	Systems administration	TBD

Future results should ideally be compared against the original Qwen3.5-4B under equivalent prompting and inference settings.

Qweble-Sol Sysadmin Evaluation

A dedicated technical evaluation is planned across areas such as:

Linux troubleshooting
macOS administration
Networking
Shell scripting
Python debugging
Docker
Server administration
DevOps
Automation
Security fundamentals

Results and evaluation prompts may be published separately for reproducibility.

Recommended Generation Settings

A reasonable starting point:

temperature: 0.6
top_p: 0.95
top_k: 20

For deterministic benchmarking, use:

temperature: 0

Context length should be chosen based on available unified memory.

Thinking / Reasoning

Qweble-Sol retains Qwen3.5's thinking-style behavior.

Depending on runtime and chat template, reasoning may appear in:

<think>
...
</think>

or as formatted reasoning before the final response.

Training Details

Qweble-Sol was fine-tuned using MLX / mlx-lm.

Base Model

Qwen/Qwen3.5-4B

Fine-tuning

Framework: mlx-lm
Method: Adapter/LoRA fine-tuning
Fine-tuner: ukuwzi

The resulting model was packaged as a 4-bit MLX release for efficient local inference.

Additional dataset and hyperparameter details may be added if released publicly.

Architecture

Qweble-Sol retains the Qwen3.5-4B architecture.

High-level properties include:

Parameters: ~4B
Hidden dimension: 2560
Transformer/hybrid blocks: 32
Feed-forward dimension: 9216
Hybrid linear-attention + full-attention architecture

Intended Uses

Qweble-Sol is primarily intended for:

Local technical assistance
Programming help
Code explanation
Debugging
Shell scripting
Systems troubleshooting
Linux/macOS administration
Docker help
Networking concepts
DevOps workflows
Automation
Educational technical assistance

It can also perform general-purpose conversational and reasoning tasks inherited from Qwen3.5-4B.

Limitations

Qweble-Sol is a relatively small ~4B-parameter model and is additionally quantized to 4-bit.

It may:

Produce incorrect technical information
Invent commands, flags, APIs, packages, or configuration options
Misinterpret ambiguous infrastructure problems
Produce code that requires additional testing
Give outdated answers about rapidly changing software
Make mistakes in long reasoning chains
Lose some accuracy compared with higher-precision versions
Incorrectly describe its own provenance without an explicit system prompt

Always verify important commands before running them.

Take extra care with:

rm
dd
disk utilities
partitioning
firewall changes
permissions
production databases
deployment commands
system configuration

Model Provenance

Item	Developer
Qwen3.5 architecture / base model	Alibaba / Qwen Team
Qwen3.5-4B base weights	Alibaba / Qwen Team
Qweble-Sol fine-tune	ukuwzi
Qweble-Sol MLX packaging	ukuwzi
Qweble-Sol GGUF conversion	ukuwzi
Qweble-Sol Ollama packaging	ukuwzi

Qweble-Sol is not an official Alibaba or Qwen release.

License

Qweble-Sol is derived from Qwen3.5-4B, which is distributed under the Apache License 2.0.

Users should review the upstream Qwen3.5 license and applicable terms before redistribution or commercial use.

Upstream base model:

Qwen/Qwen3.5-4B

Acknowledgements

Qweble-Sol builds on the work of:

Qwen Team / Alibaba — Qwen3.5-4B
MLX / mlx-lm contributors — Apple Silicon fine-tuning and inference
llama.cpp contributors — GGUF conversion and local inference
Ollama contributors — local GGUF deployment

Status

MLX 4-bit         ✅ Working
mlx-lm chat       ✅ Supported
mlx-lm generate   ✅ Supported
GGUF Q4_K_M       ✅ Available separately
llama.cpp         ✅ Tested
Ollama            ✅ Published
MMLU-Pro          ⏳ Pending
Coding benchmarks ⏳ Pending
Sysadmin eval     ⏳ Pending

Disclaimer

Qweble-Sol can generate incorrect or unsafe technical instructions. Generated commands and code should be reviewed before execution, especially on production systems or systems containing important data.

This project is an independent fine-tune and is not affiliated with or endorsed by Alibaba, Qwen, Apple, llama.cpp, or Ollama.

Downloads last month: 256

Safetensors

Model size

4B params

Tensor type

BF16

F32

MLX

Hardware compatibility

4-bit

Model tree for ukuwzi/Qweble-Sol-4B-MLX

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Finetuned

(681)

this model