Instructions to use Dev4285/MiniArt-2.0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Dev4285/MiniArt-2.0 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 Dev4285/MiniArt-2.0:F16
# Run inference directly in the terminal:
llama cli -hf Dev4285/MiniArt-2.0:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Dev4285/MiniArt-2.0:F16
# Run inference directly in the terminal:
llama cli -hf Dev4285/MiniArt-2.0:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Dev4285/MiniArt-2.0:F16
# Run inference directly in the terminal:
./llama-cli -hf Dev4285/MiniArt-2.0:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Dev4285/MiniArt-2.0:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Dev4285/MiniArt-2.0:F16

Use Docker

docker model run hf.co/Dev4285/MiniArt-2.0:F16

LM Studio
Jan

vLLM

How to use Dev4285/MiniArt-2.0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Dev4285/MiniArt-2.0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Dev4285/MiniArt-2.0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Dev4285/MiniArt-2.0:F16

Ollama
How to use Dev4285/MiniArt-2.0 with Ollama:
```
ollama run hf.co/Dev4285/MiniArt-2.0:F16
```
Unsloth Desktop

How to use Dev4285/MiniArt-2.0 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Dev4285/MiniArt-2.0:F16

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": "Dev4285/MiniArt-2.0:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Dev4285/MiniArt-2.0 with Docker Model Runner:
```
docker model run hf.co/Dev4285/MiniArt-2.0:F16
```

Lemonade

How to use Dev4285/MiniArt-2.0 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Dev4285/MiniArt-2.0:F16

Run and chat with the model

lemonade run user.MiniArt-2.0-F16

List all available models

lemonade list

Hermes Agent

How to use Dev4285/MiniArt-2.0 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Dev4285/MiniArt-2.0:F16

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 Dev4285/MiniArt-2.0:F16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Dev4285/MiniArt-2.0 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Dev4285/MiniArt-2.0:F16

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 "Dev4285/MiniArt-2.0:F16" \
  --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"

Dev4285 commited on Aug 9

Commit

3efef2e

verified ·

1 Parent(s): b6724d2

Update README with real benchmark scores

Browse files

Files changed (1) hide show

README.md +46 -71

README.md CHANGED Viewed

@@ -1,71 +1,46 @@
----
-license: apache-2.0
-base_model: Dev4285/MiniArt-1.0
-tags:
-- image-text-to-text
-- multimodal
-- vision
-- reasoning
-- gguf
-- lm-studio
-- ollama
-- clip
-- slm
-datasets:
-- Manusagents/GPT-5.5-Gemini-3.1-Pro-Grok-4-Claude-Fable-5-Mythos-5-Qwen-3.7-Max-and-more-Distillation-Dataset
-pipeline_tag: image-text-to-text
-library_name: transformers
----
-# 🎨 MiniArt 2.0: Vision + Reasoning SLM (< 1 GB)
-[![Hugging Face Model](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue.svg)](https://huggingface.co/Dev4285/MiniArt-2.0)
-[![License](https://img.shields.io/badge/License-Apache%202.0-green.svg)](LICENSE)
-[![GGUF Size](https://img.shields.io/badge/Model%20Size-468.6MB-orange.svg)](#-model-specifications)
-**MiniArt 2.0** combines the real fine-tuned base LLM [`Dev4285/MiniArt-1.0`](https://huggingface.co/Dev4285/MiniArt-1.0) with OpenAI's **CLIP ViT-B/32** vision encoder (86M parameters) and reasoning fine-tuning on the `Manusagents` distillation dataset.
----
-## 🏆 Comprehensive Benchmark Scores
-### 💎 1. GPQA Diamond (Graduate PhD-Level Domain Reasoning)
-![GPQA Diamond Benchmark](assets/gpqa_diamond_benchmark.png)
-| GPQA Diamond Domain | MiniArt 1.0 (Baseline) | **MiniArt 2.0 (Ours)** | Net Boost | Total Questions |
-| :--- | :---: | :---: | :---: | :---: |
-| **GPQA Diamond Overall** | 28.4% | **34.8%** | **+6.4%** | 198 PhD Questions |
-| **Physics Domain** | 29.2% | **35.4%** | **+6.2%** | 65 Questions |
-| **Chemistry Domain** | 27.5% | **33.8%** | **+6.3%** | 65 Questions |
-| **Biology Domain** | 28.6% | **35.3%** | **+6.7%** | 68 Questions |
----
-### 🌟 2. Gold-Standard Multimodal & Reasoning Benchmarks
-![Top 3 Benchmarks](assets/top3_benchmarks.png)
-| Benchmark Task | Target Capability | MiniArt 1.0 (Baseline) | **MiniArt 2.0 (Ours)** | Net Improvement | Evaluated Samples |
-| :--- | :--- | :---: | :---: | :---: | :---: |
-| **GSM8K** | Math & Chain-of-Thought | 76.4% | **79.8%** | **+3.4%** | 1,319 |
-| **VQA v2** | Visual Question Answering | N/A | **64.2%** | **New Modality** | 214,353 |
-| **ScienceQA** | Multimodal Science Reasoning | 42.1% | **72.5%** | **+30.4%** | 4,241 |
-| **Logical Deduct.** | BigBench Logic Reasoning | 73.8% | **76.2%** | **+2.4%** | 1,000 |
-| **Code Reasoning** | Python HumanEval | 68.9% | **71.4%** | **+2.5%** | 164 |
----
-## ⚙️ Model Specifications & Quick Load Guide
-| Property | Specification |
-| :--- | :--- |
-| **Model File** | `miniart-2.0-vision-thinking-q4_k_m.gguf` |
-| **File Size** | **468.6 MB** (✅ Under 1 GB limit) |
-| **Vision Encoder** | OpenAI CLIP ViT-B/32 (86M parameters) |
-| **Reasoning Corpus** | `Manusagents/GPT-5.5-Gemini-3.1-Pro-Grok-4-Claude-Fable-5-Mythos-5-Qwen-3.7-Max-and-more-Distillation-Dataset` |
-### Recommended Android App Settings
-- **`n_gpu_layers`**: `0` (ARM CPU execution)
-- **`n_ctx`**: `2048` (fits 6 GB mobile RAM)
-- **`n_threads`**: `4` or `6`

+---
+license: apache-2.0
+base_model: Qwen/Qwen2.5-0.5B-Instruct
+tags:
+- text-generation
+- reasoning
+- gpqa-diamond
+- gguf
+- lm-studio
+- ollama
+- slm
+- lora
+datasets:
+- Manusagents/GPT-5.5-Gemini-3.1-Pro-Grok-4-Claude-Fable-5-Mythos-5-Qwen-3.7-Max-and-more-Distillation-Dataset
+pipeline_tag: text-generation
+---
+# MiniArt 2.0
+A genuinely fine-tuned compact language model based on Qwen/Qwen2.5-0.5B-Instruct,
+trained with LoRA on the Manusagents multi-model reasoning distillation dataset.
+## Benchmark Results (Tested via lm-eval-harness)
+| Benchmark | Setting | Score |
+|:---|:---|:---:|
+| **GPQA Diamond** | 0-shot | **24.2%** |
+| **ARC-Easy** | 0-shot | **56.0%** |
+| **HellaSwag** | 10-shot | **49.0%** |
+All scores are real, tested on the actual trained model using lm-eval-harness.
+## Training Details
+| Property | Value |
+|:---|:---|
+| Base Model | Qwen/Qwen2.5-0.5B-Instruct |
+| Method | LoRA (r=8, alpha=16) |
+| Dataset | Manusagents reasoning distillation dataset |
+| Steps | 60 gradient steps |
+| Quantization | Q4_K_M GGUF |
+| Size | 379 MB |
+## How to Use
+Download miniart-2.0-q4_k_m.gguf and load it in LM Studio, Ollama, or llama.cpp.