Instructions to use unlimitedpipe/ask-0.5b-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use unlimitedpipe/ask-0.5b-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 unlimitedpipe/ask-0.5b-GGUF:Q8_0
# Run inference directly in the terminal:
llama cli -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0
# Run inference directly in the terminal:
llama cli -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0

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 unlimitedpipe/ask-0.5b-GGUF:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0

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 unlimitedpipe/ask-0.5b-GGUF:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0

Use Docker

docker model run hf.co/unlimitedpipe/ask-0.5b-GGUF:Q8_0

LM Studio
Jan

vLLM

How to use unlimitedpipe/ask-0.5b-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "unlimitedpipe/ask-0.5b-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": "unlimitedpipe/ask-0.5b-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/unlimitedpipe/ask-0.5b-GGUF:Q8_0

Ollama
How to use unlimitedpipe/ask-0.5b-GGUF with Ollama:
```
ollama run hf.co/unlimitedpipe/ask-0.5b-GGUF:Q8_0
```
Unsloth Desktop

How to use unlimitedpipe/ask-0.5b-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0

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": "unlimitedpipe/ask-0.5b-GGUF:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use unlimitedpipe/ask-0.5b-GGUF with Docker Model Runner:
```
docker model run hf.co/unlimitedpipe/ask-0.5b-GGUF:Q8_0
```

Lemonade

How to use unlimitedpipe/ask-0.5b-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unlimitedpipe/ask-0.5b-GGUF:Q8_0

Run and chat with the model

lemonade run user.ask-0.5b-GGUF-Q8_0

List all available models

lemonade list

Hermes Agent

How to use unlimitedpipe/ask-0.5b-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 unlimitedpipe/ask-0.5b-GGUF:Q8_0

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 unlimitedpipe/ask-0.5b-GGUF:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use unlimitedpipe/ask-0.5b-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unlimitedpipe/ask-0.5b-GGUF:Q8_0

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 "unlimitedpipe/ask-0.5b-GGUF:Q8_0" \
  --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"

unlimitedpipe/ask-0.5b

A 0.5B model that answers from numbered sources the way UnlimitedPipe's ask needs: it cites the sources it uses ([1]), lists every matching item when asked what is new, invents nothing, and says plainly when the sources do not cover the question, in English and Thai. Small enough (531 MB, 8-bit) to run on any machine with Ollama.

ollama pull hf.co/unlimitedpipe/ask-0.5b-GGUF
unlimited ask "any big crypto hacks this week?"      # ask picks it up by itself

unlimited setup installs it for you. This is build 4 (2026-09-27); earlier builds are on the build-3 and build-2 branches.

Scores

Questions written the way people type them ("bitcoin price?", "microsoft news", "whats new with bitget"), put through ask's own search on the public catalog of 2026-09-27 and labelled by hand: 70 in English and Thai, and 45 more in English. Every model got the same prompts, answered greedily (up to 200 new tokens, thinking off), and was graded the same way: it cites a source that answers (three of the matching ones for a list), does not cite every source, says so when the sources do not answer, invents no number, cites no source that does not exist, and answers Thai in Thai (research/ask).

Model	Parameters	Real questions (70)	More English (45)
unlimitedpipe/ask-0.5b, build 4 (this model)	0.5B	67 (95%)	45 (100%)
Qwen3.5 4B	4.2B	56 (80%)	39 (86%)
unlimitedpipe/ask-0.5b, build 3	0.5B	57 (81%)	27 (60%)
Phi-4 mini	3.8B	42 (60%)	34 (75%)
Qwen3.5 2B	1.9B	45 (64%)	29 (64%)
Llama 3.2 3B	3.2B	37 (53%)	29 (64%)
Gemma 4 E2B	5.1B (2B active)	43 (61%)
Granite 4.1 3B	3.4B	42 (60%)
LFM2.5 1.2B	1.2B	37 (53%)
unlimitedpipe/ask-0.5b, build 2	0.5B	36 (51%)
SmolLM3 3B	3.1B	35 (50%)
Qwen3.5 0.8B	0.8B	22 (31%)
Llama 3.2 1B	1.2B	21 (30%)
Gemma 3 1B	1.0B	9 (13%)
Qwen2.5 0.5B Instruct (the base)	0.5B	0 (0%)

A caution: these questions are not blind for build 4. The 70 showed where build 3 fell short (a bare topic such as "openai news" got one item), and the 45 were written after, with more of those; build 4 was made to fix exactly that. A new set of questions is the next check.

Build 4 over build 3: English first (86% of examples; Thai still passes its 19 questions), and a question as broad as a topic is answered with a list. Its misses: "nasa image of the day" gets a list where one item was wanted, and twice it answered where it should have said the sources do not cover it ("python 4 release date", "kaspa price").

Training

Qwen2.5 0.5B Instruct with LoRA (r=16, all linear layers), one pass over 17,785 examples of unlimitedpipe/ask-sft-public: the exact prompt ask sends and an answer written from the sources by templates. The sources are public data only: works of the US federal government (SEC, the Federal Register, OFAC, USGS, NOAA, NASA, CISA, FDA, DOJ, the Federal Reserve, the White House, the State Department) and sentences UnlimitedPipe writes from open data. No news articles. 147 minutes on one T4.

Limits

It answers with its sources' own words (headlines and summaries). That keeps it from inventing, and also means it does not explain or reason; a larger general model explains better.
It lists readily: a question about one thing that several sources mention gets a list.
Lists make answers longer: about 15 to 20 seconds on a 2-core CPU, against about 9 for one item.
It was taught one prompt format, ask's. Other prompts get ordinary Qwen 0.5B behaviour.
Thai answers keep English source titles as they are.

License: Apache 2.0, as its base model.

Downloads last month: -

GGUF

Model size

0.5B params

Architecture

qwen2

Hardware compatibility

8-bit

Model tree for unlimitedpipe/ask-0.5b-GGUF

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Quantized

(311)

this model

unlimitedpipe
/

ask-0.5b-GGUF

unlimitedpipe/ask-0.5b

Scores

Training

Limits

Model tree for unlimitedpipe/ask-0.5b-GGUF

Dataset used to train unlimitedpipe/ask-0.5b-GGUF