Instructions to use ANTRUhub/ANTRU-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use ANTRUhub/ANTRU-3B 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 ANTRUhub/ANTRU-3B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ANTRUhub/ANTRU-3B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ANTRUhub/ANTRU-3B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ANTRUhub/ANTRU-3B:Q4_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 ANTRUhub/ANTRU-3B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf ANTRUhub/ANTRU-3B:Q4_K_M

Use Docker

docker model run hf.co/ANTRUhub/ANTRU-3B:Q4_K_M

Llama new
LM Studio
Jan
Ollama
How to use ANTRUhub/ANTRU-3B with Ollama:
```
ollama run hf.co/ANTRUhub/ANTRU-3B:Q4_K_M
```
Unsloth Desktop

How to use ANTRUhub/ANTRU-3B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ANTRUhub/ANTRU-3B:Q4_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": "ANTRUhub/ANTRU-3B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ANTRUhub/ANTRU-3B with Docker Model Runner:
```
docker model run hf.co/ANTRUhub/ANTRU-3B:Q4_K_M
```

Lemonade

How to use ANTRUhub/ANTRU-3B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ANTRUhub/ANTRU-3B:Q4_K_M

Run and chat with the model

lemonade run user.ANTRU-3B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use ANTRUhub/ANTRU-3B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ANTRUhub/ANTRU-3B:Q4_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 ANTRUhub/ANTRU-3B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ANTRUhub/ANTRU-3B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ANTRUhub/ANTRU-3B:Q4_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 "ANTRUhub/ANTRU-3B:Q4_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"

🛑 Important Notice from the Creator

please dont make qanizations of models I make because these are very hard bulit

🙏 A note from the author

These models were hand-trained by me on a 2GB consumer laptop GPU over many long nights — they are hard to build and mean a lot to me. Please don't create quantizations or re-uploads of my models. If you build on them anyway, please clearly credit ANTRUhub and link back to the original model. Thank you.

ANTRU-3B

A 3.09B-parameter coding assistant: Qwen2.5-Coder-3B-Instruct fine-tuned with a custom curriculum on a consumer laptop (Intel i7-8550U CPU + GeForce MX150 2GB GPU, LoRA, real backprop through all adapted layers). Part of the ANTRU project — a family of fully local, self-trained models.

Model details


Base model	Qwen2.5-Coder-3B-Instruct (Apache-2.0)
Fine-tune	LoRA r=16, α=32, dropout 0.05 — all attention + MLP projections (q/k/v/o/gate/up/down)
Curriculum	302 curated examples: ANTRU identity, verified Python/coding solutions, step-by-step math reasoning, agent tool-calling format (`<tool_call>`), multi-turn clarify-then-deliver patterns
Quantization	Q4_K_M GGUF (~1.8GB) from a merged fp16 master
Context	128K tokens via YaRN (native 32K ×4)

What it's trained to do

Answer as ANTRU (built by Haider) — identity trained into the weights, not a prompt trick
Deliver complete, runnable code with explanations ("Here's a clean solution:" style)
Reason step-by-step on math and debugging questions
Emit <tool_call> JSON blocks in agent-harness format when tools are provided
Clarify once, then deliver — trained multi-turn patterns where answers to clarifying questions are followed by real code, not more questions

Honest limits

A 3B model: strong local helper, not agentic-grade — complex multi-step harness workflows (large system prompts + injected reminders) can confuse it
Trained on a small (302-example) curriculum: behavior is reliable on curriculum-like asks; broad knowledge comes from the Qwen base unchanged

Usage (llama.cpp)

llama-server -m antru-3b-Q4_K_M.gguf -c 131072 -ngl 99 \
  --rope-scaling yarn --rope-scale 4 --rope-freq-base 1000000 \
  --yarn-orig-ctx 32768 --alias antru-3b --port 8080

OpenAI-compatible endpoint at http://127.0.0.1:8080/v1 (model antru-3b). Fits a 4GB GPU at Q4; runs on CPU+RAM split for smaller cards.

Training recipe

Single epoch over 302 examples (assistant-only loss masking, ChatML framing), AdamW lr 1e-4, cosine schedule, fp32 LoRA on CPU (~2h). Merged into fp16, converted via llama.cpp convert_hf_to_gguf.py, quantized Q4_K_M. Full pipeline and ledgers in the ANTRU project repository.