Instructions to use iNiKKo/ASH-P7-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use iNiKKo/ASH-P7-4B with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="iNiKKo/ASH-P7-4B",
	filename="ASH-P7-4B-f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use iNiKKo/ASH-P7-4B 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 iNiKKo/ASH-P7-4B:F16
# Run inference directly in the terminal:
llama cli -hf iNiKKo/ASH-P7-4B:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf iNiKKo/ASH-P7-4B:F16
# Run inference directly in the terminal:
llama cli -hf iNiKKo/ASH-P7-4B: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 iNiKKo/ASH-P7-4B:F16
# Run inference directly in the terminal:
./llama-cli -hf iNiKKo/ASH-P7-4B: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 iNiKKo/ASH-P7-4B:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf iNiKKo/ASH-P7-4B:F16

Use Docker

docker model run hf.co/iNiKKo/ASH-P7-4B:F16

LM Studio
Jan

vLLM

How to use iNiKKo/ASH-P7-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "iNiKKo/ASH-P7-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "iNiKKo/ASH-P7-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/iNiKKo/ASH-P7-4B:F16

Ollama
How to use iNiKKo/ASH-P7-4B with Ollama:
```
ollama run hf.co/iNiKKo/ASH-P7-4B:F16
```

Unsloth Studio

How to use iNiKKo/ASH-P7-4B with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for iNiKKo/ASH-P7-4B to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for iNiKKo/ASH-P7-4B to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for iNiKKo/ASH-P7-4B to start chatting

How to use iNiKKo/ASH-P7-4B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf iNiKKo/ASH-P7-4B:F16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "iNiKKo/ASH-P7-4B:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use iNiKKo/ASH-P7-4B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf iNiKKo/ASH-P7-4B: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 iNiKKo/ASH-P7-4B:F16

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use iNiKKo/ASH-P7-4B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf iNiKKo/ASH-P7-4B: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 "iNiKKo/ASH-P7-4B: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"

Docker Model Runner
How to use iNiKKo/ASH-P7-4B with Docker Model Runner:
```
docker model run hf.co/iNiKKo/ASH-P7-4B:F16
```

Lemonade

How to use iNiKKo/ASH-P7-4B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull iNiKKo/ASH-P7-4B:F16

Run and chat with the model

lemonade run user.ASH-P7-4B-F16

List all available models

lemonade list

ASH-P7-4B

A QLoRA fine-tune of Qwen3-4B-Instruct-2507, trained on real pull-request and issue review discussions from Cubyz, an open-source voxel sandbox game written in Zig.

What this is

A behavior fine-tune, not a fact database. It's trained on PR/issue review judgment — real maintainer code-review reasoning (architecture critique, debugging diagnosis, "why this change is/isn't good"). It shapes how the model reasons and explains, not what it remembers.

This model is meant to be run with RAG (retrieval-augmented generation), not standalone. Facts about Cubyz should come from a retrieval pipeline that grounds answers in real docs/codebase/PR history; this fine-tune supplies the judgment and explanation style on top of that.

Accuracy on a 144-question Cubyz domain benchmark, served with RAG grounding: 99.0–99.3%.

Full project + RAG pipeline: github.com/iNiKKo/ASH-AI

Files in this repo

File	What it is
`ASH-P7-4B-q4.gguf`	Merged model, Q4 quantized (~2.4GB) — ready to run directly with Ollama or llama.cpp.
`ASH-P7-4B-f16.gguf`	Merged model, full precision (~7.5GB) — for anyone who wants to requantize themselves.
`adapter_model.safetensors` + `adapter_config.json`	The raw LoRA adapter (~127MB) — apply this to the base model yourself instead of using a pre-merged GGUF.

Quick start (Ollama)

``` ollama create ASH-P7-4B -f Modelfile ``` with a Modelfile containing: ``` FROM ./ASH-P7-4B-q4.gguf PARAMETER num_ctx 16384 ```

License

The base model (Qwen3-4B-Instruct-2507) is Apache 2.0, and this repo is labeled Apache 2.0 to match. One thing worth knowing: the fine-tuning data is derived from PR/issue review discussions on Cubyz, which is itself GPL-3.0 licensed. Whether that meaningfully affects the license status of model weights trained on that data (as opposed to the underlying source code itself) is genuinely unsettled territory, not something this repo resolves for you — flagged here so you have the full picture rather than an unqualified license claim.

Downloads last month: -

GGUF

Model size

4B params

Architecture

qwen3

Hardware compatibility

16-bit

View +1 variant

Model tree for iNiKKo/ASH-P7-4B

Base model

Qwen/Qwen3-4B-Instruct-2507

Adapter

(5636)

this model