Instructions to use TheZeez/gemma-4-e4b-creative-DFT-exp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use TheZeez/gemma-4-e4b-creative-DFT-exp with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="TheZeez/gemma-4-e4b-creative-DFT-exp", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("TheZeez/gemma-4-e4b-creative-DFT-exp")
model = AutoModelForMultimodalLM.from_pretrained("TheZeez/gemma-4-e4b-creative-DFT-exp", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use TheZeez/gemma-4-e4b-creative-DFT-exp with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "TheZeez/gemma-4-e4b-creative-DFT-exp"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "TheZeez/gemma-4-e4b-creative-DFT-exp",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/TheZeez/gemma-4-e4b-creative-DFT-exp

SGLang

How to use TheZeez/gemma-4-e4b-creative-DFT-exp with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "TheZeez/gemma-4-e4b-creative-DFT-exp" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "TheZeez/gemma-4-e4b-creative-DFT-exp",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "TheZeez/gemma-4-e4b-creative-DFT-exp" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "TheZeez/gemma-4-e4b-creative-DFT-exp",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Unsloth Studio

How to use TheZeez/gemma-4-e4b-creative-DFT-exp 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 TheZeez/gemma-4-e4b-creative-DFT-exp 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 TheZeez/gemma-4-e4b-creative-DFT-exp to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for TheZeez/gemma-4-e4b-creative-DFT-exp to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="TheZeez/gemma-4-e4b-creative-DFT-exp",
    max_seq_length=2048,
)

Docker Model Runner
How to use TheZeez/gemma-4-e4b-creative-DFT-exp with Docker Model Runner:
```
docker model run hf.co/TheZeez/gemma-4-e4b-creative-DFT-exp
```

gemma-4-e4b-creative-DFT-exp

File size: 1,750 Bytes

fea339a
 
 
 
 
 
 
 
 
 
 
9acf3f0
16a1e4f

---
base_model: coder3101/gemma-4-E4B-it-heretic
tags:
- text-generation-inference
- transformers
- unsloth
- gemma4
license: apache-2.0
language:
- en
---
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)

*goated software^^*

# DISCLAIMER

**I am not a mathematician nor a professional coder. This was an experiment (with the help of AI of course).**

This is a custom-trained version of Google's Gemma 4 E4B intended for creative writing. It was trained using a custom implementation of **Distribution Fine-Tuning (DFT)** designed to mathematically penalize and eliminate repetitive AI slop and predictable phrasing.

The core training algorithm was inspired by the concepts outlined in the May 18, 2026 blog post, [*Fixing LLM Writing with Distribution Fine-Tuning*](https://rosmine.ai/2026/05/18/fixing-llm-writing-with-distribution-fine-tuning/). 

Standard Supervised Fine-Tuning (SFT) and RLHF often cause models to regress to a generic, hyper-structured average of human text. To counter this, this model was trained by injecting a macro-statistical loss penalty into the backpropagation loop. By calculating the Mean Squared Error (MSE) between the model's batch-level vocabulary distribution and a human target distribution (or a good creative dataset), the model was actively penalized for overusing AI-frequent vocabulary (e.g., "whisper", "shiver", "sheer").

The model was trained on 1.3~ Epochs and effective batch size of 96, using a mix of multiturn roleplaying and creative writing dataset.

Credits to Rosmine and Google Gemini for the idea and the implementation. Let me know what you think in the Community section!