Gemma 3 4b Abliterated

🚀 Gemma 3 4B (Abliterated Text-Only) Model Card

This is a text-only model that solves the problem of providing a text - only alternative to the multimodal Gemma - 3 - 4b model. It allows users to focus solely on text generation without the need for vision - related components.

🚀 Quick Start

You can load and use this model the same way you would use the text - only google/gemma-3-1b-it version. The following is a code example to show you how to load and use this model:

from transformers import AutoTokenizer, BitsAndBytesConfig, Gemma3ForCausalLM
import torch

model_id = "gghfez/gemma-3-4b-novision"

quantization_config = BitsAndBytesConfig(load_in_8bit=True)

model = Gemma3ForCausalLM.from_pretrained(
    model_id, quantization_config=quantization_config
).eval()

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    [
        {
            "role": "system",
            "content": [{"type": "text", "text": "You are a helpful assistant."},]
        },
        {
            "role": "user",
            "content": [{"type": "text", "text": "Write a poem on Hugging Face, the company"},]
        },
    ],
]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device).to(torch.bfloat16)


with torch.inference_mode():
    outputs = model.generate(**inputs, max_new_tokens=64)

outputs = tokenizer.batch_decode(outputs)

✨ Features

• Text - Only Version: This is an abliterated text - only version of google/gemma-3-4b-it, created using Baukit.
• Modified Architecture: It uses the same architecture as the text - only 1b model, with the vision components removed.
• Large Parameter Count: It has 4 billion parameters, which can potentially provide rich text - generation capabilities.

📚 Documentation

Model Description

Important Note

This model was created by gghf by removing the vision encoders. Please note that this model may exhibit a reduced performance.

License

The license for this model is gemma.