Instructions to use devopsfortunato/Fonzie0.2-4B-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 devopsfortunato/Fonzie0.2-4B-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 devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf devopsfortunato/Fonzie0.2-4B-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 devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf devopsfortunato/Fonzie0.2-4B-GGUF: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 devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M

Llama new
LM Studio
Jan
Ollama
How to use devopsfortunato/Fonzie0.2-4B-GGUF with Ollama:
```
ollama run hf.co/devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M
```
Unsloth Desktop

How to use devopsfortunato/Fonzie0.2-4B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf devopsfortunato/Fonzie0.2-4B-GGUF: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": "devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use devopsfortunato/Fonzie0.2-4B-GGUF with Docker Model Runner:
```
docker model run hf.co/devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M
```

Lemonade

How to use devopsfortunato/Fonzie0.2-4B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Fonzie0.2-4B-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use devopsfortunato/Fonzie0.2-4B-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 devopsfortunato/Fonzie0.2-4B-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 devopsfortunato/Fonzie0.2-4B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use devopsfortunato/Fonzie0.2-4B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf devopsfortunato/Fonzie0.2-4B-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 "devopsfortunato/Fonzie0.2-4B-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"

Fonzie0.2-4B-GGUF

GGUF files of Fonzie0.2-4B: a Qwen3.5-4B fine-tune that searches when it doesn't know, answers with sources, speaks Italian and English, and looks at pictures. It is trained to work inside agent apps (oh-my-pi, opencode). Compared to Fonzie0.1 it tries again after a failed search, searches first for "the latest X", says when something doesn't exist, and asks the user with the ask tool in oh-my-pi. Evaluation, limits and training details are in the model card of the full model, Fonzie0.2-4B.

File	Size	Use
`Fonzie0.2-4B-Q8_0.gguf`	4.6 GB	closest to the bf16; not evaluated yet
`Fonzie0.2-4B-Q4_K_M.gguf`	2.8 GB	smaller and faster; evaluated against the bf16 below
`mmproj-Fonzie0.2-4B-F16.gguf`	0.7 GB	vision projector, needed for pictures

Made with llama.cpp (convert_hf_to_gguf.py, llama-quantize). The Q4_K_M is quantized from an F16 conversion. The evaluations in the model card of Fonzie0.2-4B used the bf16 model; the Q4_K_M was evaluated separately, below.

Memory by context

GPU memory used by llama-server with the vision projector loaded, MTP on and the default f16 KV cache:

Context (tokens)	Q4_K_M	Q8_0
8,192	4.2 GiB	5.9 GiB
32,768	5.1 GiB	6.8 GiB
65,536	6.3 GiB	8.0 GiB
131,072	8.7 GiB	10.4 GiB
163,840	10.1 GiB	11.8 GiB
262,144 (the model's limit)	≥ 13.4 GiB	≥ 15.1 GiB

Measured: the Q4_K_M on an RTX 4070 Ti (12 GB) at 64k, 128k and 160k, the last two with a prompt as long as the context.
Computed: the other Q4_K_M rows are the measured fixed part (3.9 GiB) plus about 38 KB per token. The Q8_0 rows add its bigger file (1.7 GiB) to the Q4_K_M; the Q8_0 itself was not run.
q8_0 KV cache (-ctk q8_0 -ctv q8_0): less than half the memory per token. On the 12 GB card the Q4_K_M then fits 229,376 tokens with the projector (10.0 GiB at peak), or the full 262,144 for text only (10.0 GiB).
Without MTP and without the projector the fixed part drops to 3.0 GiB.
Desktop apps hold memory on the same card: leave a margin.

How much the Q4_K_M loses

The same suites as the model card, Q4_K_M on llama-server against bf16 on vLLM 0.32. Bold marks the differences whose paired 95% interval excludes zero.

	bf16	Q4_K_M
frozen test: accuracy	66.5%	64.8%
frozen test: confident wrong answers	24.6%	27.4%
frozen test: invented links	2.6%	7.2%
everyday questions, accuracy (oh-my-pi / opencode / Hermes)	66.9 / 66.7 / 76.4%	54.2 / 66.7 / 73.2%
new test set, accuracy (oh-my-pi / opencode / Hermes)	55.4 / 61.1 / 67.8%	52.1 / 67.2 / 60.6%

In opencode and Hermes the Q4_K_M is as good as the bf16, within the noise. In oh-my-pi it loses about 13 points on everyday questions, and it writes more links that no tool returned. These runs can't separate the 4-bit weights from the engine (llama.cpp instead of vLLM).

Speed on an RTX 4070 Ti, one user: 118.1 tokens/s, 157.0 with MTP 2 (the model's own multi-token prediction head, used for speculative decoding).

Usage

LM Studio: load the model file together with the mmproj file.

llama.cpp:

llama-server -m Fonzie0.2-4B-Q4_K_M.gguf --mmproj mmproj-Fonzie0.2-4B-F16.gguf --jinja -c 65536 \
  --temp 1.0 --top-p 0.95 --top-k 20 --min-p 0

A 12 GB card, the fastest setup measured (llama.cpp 876a43211), with MTP 2 and a q8_0 KV cache:

llama-server -m Fonzie0.2-4B-Q4_K_M.gguf --mmproj mmproj-Fonzie0.2-4B-F16.gguf -ngl 99 -np 1 \
  -c 229376 -ctk q8_0 -ctv q8_0 --spec-type draft-mtp --spec-draft-n-max 2 \
  --jinja --reasoning-format deepseek --temp 1.0 --top-p 0.95 --top-k 20 --min-p 0

Give it a web search tool: without tools and a system prompt it falls back to the base model's long thinking.

License: Apache 2.0, like the base model Qwen3.5-4B.

Author: Alfonso Fortunato.

Downloads last month: -

GGUF

Model size

4B params

Architecture

qwen35

Hardware compatibility

4-bit

8-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for devopsfortunato/Fonzie0.2-4B-GGUF

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Quantized

(602)

this model