Instructions to use NAME0x0/AVA-v2-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NAME0x0/AVA-v2-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="NAME0x0/AVA-v2-GGUF",
	filename="AVA-v2-IQ4_XS.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 NAME0x0/AVA-v2-GGUF 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 NAME0x0/AVA-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf NAME0x0/AVA-v2-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf NAME0x0/AVA-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf NAME0x0/AVA-v2-GGUF:Q4_K_M

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 NAME0x0/AVA-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf NAME0x0/AVA-v2-GGUF:Q4_K_M

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 NAME0x0/AVA-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf NAME0x0/AVA-v2-GGUF:Q4_K_M

Use Docker

docker model run hf.co/NAME0x0/AVA-v2-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use NAME0x0/AVA-v2-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NAME0x0/AVA-v2-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NAME0x0/AVA-v2-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/NAME0x0/AVA-v2-GGUF:Q4_K_M

Ollama
How to use NAME0x0/AVA-v2-GGUF with Ollama:
```
ollama run hf.co/NAME0x0/AVA-v2-GGUF:Q4_K_M
```

Unsloth Studio

How to use NAME0x0/AVA-v2-GGUF 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 NAME0x0/AVA-v2-GGUF 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 NAME0x0/AVA-v2-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for NAME0x0/AVA-v2-GGUF to start chatting

How to use NAME0x0/AVA-v2-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NAME0x0/AVA-v2-GGUF:Q4_K_M

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": "NAME0x0/AVA-v2-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use NAME0x0/AVA-v2-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NAME0x0/AVA-v2-GGUF: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 NAME0x0/AVA-v2-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use NAME0x0/AVA-v2-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NAME0x0/AVA-v2-GGUF: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 "NAME0x0/AVA-v2-GGUF: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"

Docker Model Runner
How to use NAME0x0/AVA-v2-GGUF with Docker Model Runner:
```
docker model run hf.co/NAME0x0/AVA-v2-GGUF:Q4_K_M
```

Lemonade

How to use NAME0x0/AVA-v2-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull NAME0x0/AVA-v2-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.AVA-v2-GGUF-Q4_K_M

List all available models

lemonade list

AVA logo

AVA v2 — GGUF

Ready-to-run GGUF builds of AVA v2, a 2B reasoning model fine-tuned entirely on a single 4 GB laptop GPU. 82.0% ARC-Challenge, 92.0% ARC-Easy, 59.2% MMLU on a 17-benchmark / 16,872-task full evaluation (report).

Works with llama.cpp, Ollama, LM Studio, Jan, KoboldCpp — no Python, no GPU required.

Files

All sub-8-bit quants are built with an importance matrix calibrated on the model's own training distribution (reasoning, math, science, instruction following) — the same idea behind Google's Gemma QAT releases: keep the small quants as close to reference quality as possible.

Measured quality cost vs the Q8_0 reference (perplexity on a held-out slice of the training distribution, 512-token context — lower is better):

File	Size	RAM needed	PPL	vs Q8_0	Use when
AVA-v2-IQ4_XS.gguf	1.11 GB	~1.6 GB	2.5347	+2.0%	Tightest fit — old laptops, SBCs
AVA-v2-Q4_0.gguf	1.12 GB	~1.6 GB	2.5244	+1.6%	ARM/AVX-optimized CPU inference
AVA-v2-Q4_K_M.gguf	1.19 GB	~1.7 GB	2.4907	+0.25%	Recommended default
AVA-v2-Q5_K_M.gguf	1.31 GB	~1.8 GB	—	—	Better quality, still small
AVA-v2-Q8_0.gguf	1.87 GB	~2.4 GB	2.4844	reference	Matches the published eval

Quick start

Ollama

ollama run hf.co/NAME0x0/AVA-v2-GGUF:Q4_K_M

llama.cpp

llama-cli -m AVA-v2-Q4_K_M.gguf -ngl 99 --temp 0.7 \
  -p "Explain why ice floats on water."

LM Studio / Jan

Search for NAME0x0/AVA-v2-GGUF in the model browser and download a file.

Chat format

Qwen3.5 ChatML-style template (embedded in the GGUF — runtimes apply it automatically):

<|im_start|>user
{your prompt}<|im_end|>
<|im_start|>assistant

Benchmarks (Q8_0, full sets, 95% Wilson CI)

Benchmark	n	Accuracy
ARC-Easy	2,376	92.0%
ARC-Challenge	1,172	82.0%
PIQA	1,838	75.9%
BoolQ	3,270	75.0%
MMLU (5-shot)	14,042	59.2%
GSM8K (greedy / k=5)	1,319 / 200	35.3% / 44.0%

Full 17-benchmark table and protocol: RESULTS_REPORT_V2_FULL.md.

At 2B parameters, AVA v2's ARC-Challenge (82.0%) sits ahead of Llama 3.2 3B-Instruct (78.6%) and within two points of Phi-4-mini 3.8B (83.7%) — models trained with cluster-scale compute. AVA v2 was trained in 100 minutes on one 4 GB laptop GPU.

Provenance

Adapter + training details: NAME0x0/AVA-v2
Base model: Qwen/Qwen3.5-2B (Apache 2.0)
Everything reproducible: github.com/NAME0x0/AVA — corpus builders, training configs, eval harness, and this quantization pipeline are all in the repo.

Citation

@misc{ava-v2-2026,
  title={AVA v2: QLoRA Fine-tuning Under Extreme VRAM Constraints},
  author={Muhammad Afsah Mumtaz},
  year={2026},
  url={https://github.com/NAME0x0/AVA}
}