Instructions to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M
# Run inference directly in the terminal:
llama cli -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M
# Run inference directly in the terminal:
llama cli -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Use Docker

docker model run hf.co/M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Llama new
LM Studio
Jan

vLLM

How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "M1n1A1/MiniAI-Quata1.5-Reasoning-4b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "M1n1A1/MiniAI-Quata1.5-Reasoning-4b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Ollama
How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with Ollama:
```
ollama run hf.co/M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M
```
Unsloth Desktop

How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with Docker Model Runner:
```
docker model run hf.co/M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M
```

Lemonade

How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Run and chat with the model

lemonade run user.MiniAI-Quata1.5-Reasoning-4b-Q5_K_M

List all available models

lemonade list

Hermes Agent

How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

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 M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use M1n1A1/MiniAI-Quata1.5-Reasoning-4b with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M

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 "M1n1A1/MiniAI-Quata1.5-Reasoning-4b:Q5_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MiniAI Quata1.5 Reasoning (4B)

Big-model quality. 4 billion parameters. Now with native chain-of-thought reasoning.

Quata1.5 Reasoning is the reasoning-tuned evolution of MiniAI Quata1.5 — the same compact, high-performing 4B language model built on a Qwen3-4B foundation, enhanced with native Qwen3 thinking tokens and a dedicated chain-of-thought fine-tune. It doesn't just answer — it thinks step by step before it speaks, delivering coherent, verifiable reasoning on math, logic, code, science, and everyday questions, all in a package that fits on a single modest GPU or runs entirely on your own hardware.

Highlights

Native chain-of-thought — emits Qwen3 <|thinking|> / <|answer|> style reasoning tokens
2.9 GB GGUF (Q5_K_M) — fast on consumer hardware, easy to self-host
100% on-device option — nothing leaves your machine
Reasoning where it counts — see the benchmarks below

Benchmarks

Quata1.5 Reasoning holds its own against models many times its size.

MMLU — knowledge across 30+ subjects

ARC-Challenge — science reasoning

GSM8K — grade school math

TruthfulQA — truthfulness

HellaSwag — commonsense reasoning

Model	MMLU	ARC-Challenge	GSM8K	TruthfulQA	HellaSwag
GPT-4 (~1.8T)	86.4	-	-	-	-
Gemma 3 (4B)	84.4	88.0%	100.0%	92.0%	100.0%
MiniAI Quata1.5 (4B)	84.4	92.0%	96.0%	88.0%	72.0%
MiniAI Quata1.5 Reasoning (4B)	88.9	100.0%	92.0%	92.0%	96.0%

Gaining the ability to think before it answers, Quata1.5 Reasoning jumps to a 93.8% average across the suite — +7 points over the base Quata1.5, matching or beating a 4B-class leader on MMLU, ARC-Challenge, and TruthfulQA, and landing within a hair of models with over 400× its parameters. That's the return you get from a tuned, focused 4B model that reasons carefully instead of pattern-matching.

Get started

Run it locally (Ollama)

ollama run hf.co/M1n1A1/MiniAI-Quata1.5-Reasoning-4b:GGUF

Or pull it explicitly:

ollama pull hf.co/M1n1A1/MiniAI-Quata1.5-Reasoning-4b:GGUF

Hosted API

Soon™

Details

Architecture: Qwen3-based, 4B parameters
Reasoning: native Qwen3 thinking tokens, chain-of-thought fine-tuned
Quantization (GGUF): Q5_K_M (~2.9 GB)
Context length: 40960 tokens, 36 layers, hidden size 2560
License: Apache 2.0 (base)
More: (Removed)

Downloads last month: 52

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for M1n1A1/MiniAI-Quata1.5-Reasoning-4b

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Quantized

(364)

this model

Quantizations

2 models