Instructions to use agentscope-ai/Piccolo-4B-Q8_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 agentscope-ai/Piccolo-4B-Q8_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 agentscope-ai/Piccolo-4B-Q8_0:Q8_0
# Run inference directly in the terminal:
llama cli -hf agentscope-ai/Piccolo-4B-Q8_0:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf agentscope-ai/Piccolo-4B-Q8_0:Q8_0
# Run inference directly in the terminal:
llama cli -hf agentscope-ai/Piccolo-4B-Q8_0:Q8_0

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 agentscope-ai/Piccolo-4B-Q8_0:Q8_0
# Run inference directly in the terminal:
./build/bin/llama cli -hf agentscope-ai/Piccolo-4B-Q8_0:Q8_0

Use Docker

docker model run hf.co/agentscope-ai/Piccolo-4B-Q8_0:Q8_0

Llama new
LM Studio
Jan
Ollama
How to use agentscope-ai/Piccolo-4B-Q8_0 with Ollama:
```
ollama run hf.co/agentscope-ai/Piccolo-4B-Q8_0:Q8_0
```
Unsloth Desktop

How to use agentscope-ai/Piccolo-4B-Q8_0 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf agentscope-ai/Piccolo-4B-Q8_0: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": "agentscope-ai/Piccolo-4B-Q8_0:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use agentscope-ai/Piccolo-4B-Q8_0 with Docker Model Runner:
```
docker model run hf.co/agentscope-ai/Piccolo-4B-Q8_0:Q8_0
```

Lemonade

How to use agentscope-ai/Piccolo-4B-Q8_0 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull agentscope-ai/Piccolo-4B-Q8_0:Q8_0

Run and chat with the model

lemonade run user.Piccolo-4B-Q8_0-Q8_0

List all available models

lemonade list

Hermes Agent

How to use agentscope-ai/Piccolo-4B-Q8_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 agentscope-ai/Piccolo-4B-Q8_0: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 agentscope-ai/Piccolo-4B-Q8_0:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use agentscope-ai/Piccolo-4B-Q8_0 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf agentscope-ai/Piccolo-4B-Q8_0: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 "agentscope-ai/Piccolo-4B-Q8_0: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"

Piccolo-4B-Q8_0

GitHub · 🌐 Decision · Hugging Face · ModelScope · coming soon

Q8_0 GGUF of Piccolo-4B, for automatic tool review and context compaction in QwenPaw. The complete original Safetensors checkpoint and reference inference code are kept in the base repository.

Download and enable

In QwenPaw's local model list, download agentscope-ai/Piccolo-4B-Q8_0 and click Start. Confirm the default Decision model, then enable Decision Model Mode for tool review and Decision model compaction in their existing settings. Starting Piccolo does not enable either feature and the local runtime serves one model at a time.

Piccolo requires the companion llama.cpp patch in QwenPaw PR #8463. An unpatched causal chat runtime or ordinary Ollama Chat Completions does not implement this model. There is no additional service URL to configure.

Runtime and quantization

The text backbone was converted to BF16 GGUF with the patched converter using --no-mtp, then quantized with llama-quantize ... Q8_0. All text backbone layers and the output head are retained. The vision encoder and auxiliary MTP tensors are not used in this text decision path; they remain in the complete original checkpoint. No importance matrix or calibration dataset was used for this quantization.

The runtime uses the checkpoint's exact flex_twice input layout, bidirectional prompt attention in full-attention layers, and unchanged linear-attention computation. It reads single-token candidate logits at Answer: ( and applies candidate softmax at temperature 1, generating no answer tokens. Disable prefix-cache reuse, chunked prefill, and silent truncation.

All GGUF shards in this repository are required. release-manifest.json records their SHA-256 hashes and sizes. The validation artifact covers 4 smoke questions against the complete checkpoint; maximum absolute candidate probability difference was 0.00564096, with matching top choices. This is numerical smoke evidence, not benchmark, safety, or calibration validation. Preserve QwenPaw's native approval and compaction fallback.

Release entry points

The model matrix lists all 12 published full/BF16/Q8_0/Q4_K_M repositories. Full-checkpoint benchmark results are separate from the conversion/quantization measurements in this card; no identical quantized benchmark scores are implied. ModelScope is not published yet. QwenPaw integration remains in Draft PR #8463; use the pinned setup guide and its companion patched runtime.