Instructions to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF",
	filename="Gemma4-E2B-it-Heretic_BF16.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-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 Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-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 Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with Ollama:
```
ollama run hf.co/Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M
```

Unsloth Studio new

How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-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 Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-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 Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF to start chatting

Pi new

How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-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": "Gemma4-E2B-it-Uncensored-Alienstro-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with Docker Model Runner:
```
docker model run hf.co/Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M
```

Lemonade

How to use Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Gemma4-E2B-it-Uncensored-Alienstro-GGUF-Q4_K_M

List all available models

lemonade list

🦙 Gemma4-E2B-it-Heretic GGUF

This repository contains GGUF quants for Gemma4-E2B-it-Heretic, an uncensored version of Google's Gemma 4 E2B it. This model is specifically optimized for high instruction-following compliance and reduced safety-filter interference, aimed at providing more direct and unrestricted responses.

📊 Performance Metrics

Refusal Rate: 7/100 (7%). In internal testing, the model successfully followed instructions in 93% of scenarios where base models typically decline due to safety buffers.
Training Note: KL-Divergence (Kullback–Leibler divergence) metrics for this specific pass were not recorded because I forgot!

🚀 Benchmarks

The following benchmarks were conducted on an NVIDIA GeForce RTX 5060 Ti (16GB) using the llama-bench tool with Vulkan offloading and 8 threads.

Model Variant	Test	Context	Tokens/sec (t/s)
Q4_K_M (Medium)	Prompt Processing	512 tokens	6804.99 ± 261.99
Q4_K_M (Medium)	Text Generation	128 tokens	158.50 ± 1.09
Q8_0	Prompt Processing	512 tokens	7439.29 ± 730.78
Q8_0	Text Generation	128 tokens	118.76 ± 0.17

🗜️ Files & Quantization

All quants were generated using a storage-aware llama.cpp pipeline.

🛠️ Usage

Llama.cpp

-cnv: Conversation mode. It manages chat templates so the model acts like an assistant instead of just completing text.

-ngl: GPU Offloading. "Number of GPU Layers." Setting this to 99 puts the entire model on your graphics card for maximum speed.

-c: Context Size. The model's "short-term memory." High values (like 4096) allow longer chats but use more VRAM.

-t: CPU Threads. The number of CPU cores used to process any parts of the model that didn't fit on the GPU.

--mmap: Memory Mapping. Efficiently uses Free RAM by reading only the parts of the model file needed at that moment.

--no-mlock: Unlock Memory. Prevents the model from "locking" into physical RAM. This keeps your system stable by letting the OS move data if RAM gets low.

You can run these models using the llama-cli built during the pipeline:

./llama-cli -m Gemma4-E2B-it-Heretic_Q4_K_M.gguf -cnv -ngl 99 -c 4096 -t 8 --mmap --no-mlock

Downloads last month: 2,564

GGUF

Model size

5B params

Architecture

gemma4

Hardware compatibility

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Inference Providers NEW

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

Model tree for Alienstro/Gemma4-E2B-it-Uncensored-Alienstro-GGUF

Base model

google/gemma-4-E2B

Finetuned

google/gemma-4-E2B-it

Quantized

(171)

this model