Instructions to use AtomicChat/gemma-4-E4B-it-assistant-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 AtomicChat/gemma-4-E4B-it-assistant-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 AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf AtomicChat/gemma-4-E4B-it-assistant-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 AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf AtomicChat/gemma-4-E4B-it-assistant-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 AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf AtomicChat/gemma-4-E4B-it-assistant-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 AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M

Use Docker

docker model run hf.co/AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use AtomicChat/gemma-4-E4B-it-assistant-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AtomicChat/gemma-4-E4B-it-assistant-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AtomicChat/gemma-4-E4B-it-assistant-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M

Ollama
How to use AtomicChat/gemma-4-E4B-it-assistant-GGUF with Ollama:
```
ollama run hf.co/AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use AtomicChat/gemma-4-E4B-it-assistant-GGUF with Docker Model Runner:
```
docker model run hf.co/AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M
```

Lemonade

How to use AtomicChat/gemma-4-E4B-it-assistant-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull AtomicChat/gemma-4-E4B-it-assistant-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.gemma-4-E4B-it-assistant-GGUF-Q4_K_M

List all available models

lemonade list

Atomic Chat

How can I run mmproj + draft acceleration together on the model?

by VaLtEc-BoY - opened May 10

Discussion

VaLtEc-BoY

May 10

Hello, how can I run mmproj + draft acceleration together on the model? For example, running GEMMA-4-E4B-gguf + draft assistant + mmproj (vision) + turbo quant Q3. Tks! I like the project atomic-llama-cpp-turboquant !!

Biogenic

May 13

Hey, thanks for the kind words! 🙏

Unfortunately, mmproj + speculative decoding cannot run together — llama-server force-disables any draft pipeline as soon as a multimodal projector is loaded (image chunks expand to multiple KV tokens and the draft path doesn't handle them yet). This is an upstream llama.cpp limitation, not specific to this fork.

You can pick any two of three:

Vision + turbo3 KV — drop --mtp-head
Draft + turbo3 KV — drop --mmproj (this is what our scripts/run-gemma4-e4b-mtp-server.sh does)
We're putting mmproj + draft support on the roadmap and will look into it — it needs the draft pipeline to step over mtmd image chunks. Will share updates once we have something to test!

Biogenic

May 13

Quick update — turned out simpler than expected, just merged it: https://github.com/AtomicBot-ai/atomic-llama-cpp-turboquant/pull/14

You can now load --mmproj together with --spec-type mtp / nextn / eagle3 on the same llama-server, so your GEMMA-4-E4B + assistant + mmproj + turbo3 combo works out of the box.

Small honest caveat: draft acceleration kicks in on text-only turns. When a turn actually contains an image, that turn falls back to plain target decoding (image still recognised correctly). Lifting that for vision turns is on the follow-up list.

Tested on M4 Max with Qwen 3.6 + NextN and Gemma 4 + MTP — both happy. Enjoy! 🙌

VaLtEc-BoY

May 13

Thank you very much for your reply, your project is very good. I've been using Kobold, but I need maximum efficiency. Turbo Quant will help me a lot... I believe that Llama, or any version of it, is important, especially because I don't want anything online, which is what I'm currently aiming for!! Offline agents and chats, your project is perfect for that!TKS!!

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