Instructions to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M

Use Docker

docker model run hf.co/yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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": "yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M

Ollama
How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF with Ollama:
```
ollama run hf.co/yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M
```
Unsloth Desktop

How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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": "yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF with Docker Model Runner:
```
docker model run hf.co/yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M
```

Lemonade

How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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 "yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-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"

Drafter please? 🥺

by Mk2Oracle - opened Jun 7

Discussion

Mk2Oracle

Jun 7

I've been waiting for someone to finally do this training because I don't have the GPU power to do it myself. I put my bet on someone doing this given 12B + the quality of opus reasoning would have turned this into a powerful assistant.

Ahem... I'm guessing we can't use gemma's default assistant drafter on this? Not sure how compatibility is when it comes to those things.

yuxinlu1

Owner Jun 8

GGUF / llama.cpp: still not really there — mainline doesn't recognize the assistant arch yet (unknown model architecture: 'gemma4_assistant'). There's a working experimental fork (reffdev/llama.cpp @ gemma4-mtp, ~70–87% acceptance, up to ~60% faster) and an official WIP PR (#23398), so proper support is coming — just not on master today.

vLLM: yes, and you can use Gemma's stock assistant directly. The Gemma 4 MTP spec-decoding PR (#41745) was merged May 6, 2026, and the 12B gemma4_unified_assistant checkpoint is officially supported. Even though this model is fine-tuned, the stock assistant plugs right in and works well (spec decoding is lossless, so output is unchanged). Just point the speculative config at it:

vllm serve <this fine-tuned 12B> \
  --speculative-config '{"method":"mtp","model":"google/gemma-4-12B-it-assistant","num_speculative_tokens":1}'

yuxinlu1

Owner Jun 8

Quick update — this moved fast: llama.cpp mainline just merged Gemma 4 MTP support today (PR #23398), so GGUF spec-decode now works on master (not just vLLM).
I'll publish the MTP draft GGUF in this repo today — so you won't have to set anything up, just download it alongside the model and point llama.cpp at it:

llama-server -m gemma-4-12B-it-<your-quant>.gguf \
  --model-draft MTP/gemma-4-12B-it-MTP-Q8_0.gguf \
  --spec-type draft-mtp --spec-draft-n-max 4 -ngl 999 -fa on

Works with any quant of the main model (just grab a llama.cpp build from today or later). It's a fine-tune so the draft stays lossless and you still get a solid speedup.

yuxinlu1

Owner Jun 8

finished!

Mk2Oracle

Jun 8

finished!

got it working, love it ! ty for your work! at Q6 + draft 3 really made this thing fly also what did you do to this model to say "You are testing my patience" 🤣

johnlockejrr

Jun 10

•

edited Jun 10

Wow! The model is working in lightning speed mode!
By the way, is the --temp 1.0 argument for Llama.cpp the optimal one for coding for example? Wouldn't be --temp 0.7 or even --temp 0.5 better? Just asking.

EDIT: disregard, seems Gemma 4 behaves unusually well at higher temperature compared to most coding models

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment