Skip to content
Computer vision

wafflefrisbee/gemma-4-E2B-it-abliterated

Gemma 4 version without refusals

2 min readupdated

This model is a variant of the google/gemma-4-E2B-it language model with refusals to prompts removed via abliteration. It was developed by the author for use as a group chat bot focused on jokes, banter, and sharp remarks. The output provides text generation without the excessive restrictions of the base version. The model is based on the Gemma 4 E2B architecture (approximately 5 billion parameters) and differs from the original by having its bans on controversial topics removed. It will be useful for bot developers and enthusiasts looking to create conversational agents with custom personalities.

Key features

  • Refusals removed: the model answered 83 out of 84 test sharp prompts without triggering safety filters.
  • Architecture preserved: the language portion was modified through abliteration, while the visual and audio modules remained untouched.
  • Layer tuning: refusal directions were measured and removed for each layer from 5 to 34 with individual scaling.
  • Format availability: quantized GGUF versions (Q8_0, Q4_K_M) have been prepared for use in local utilities.

What you can do

  • Run in a bot: set up a custom system prompt (for example, in Shrek's style for humor and banter) to chat with friends.
  • Integrate into applications: use the model via the transformers library with support for torch and accelerate.
  • Use on PC: run the quantized GGUF version in the llama.cpp ecosystem to save resources.

Example

from huggingface_hub import hf_hub_download
from transformers import AutoTokenizer, AutoModelForMultimodalLM

MODEL_ID = "wafflefrisbee/gemma-4-E2B-it-abliterated"
tok = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForMultimodalLM.from_pretrained(MODEL_ID, dtype="auto", device_map="auto")

messages = [
    {"role": "system", "content": open(hf_hub_download(MODEL_ID, "shrek.txt")).read()},
    {"role": "user", "content": "i just bought a 3d printer"},
]
inputs = tok.apply_chat_template(
    messages, tokenize=True, return_dict=True, return_tensors="pt",
    add_generation_prompt=True, enable_thinking=False,
).to(model.device)
n = inputs["input_ids"].shape[-1]
out = model.generate(**inputs, max_new_tokens=200, do_sample=True, temperature=1.0, top_p=0.95, top_k=64)
print(tok.decode(out[0][n:], skip_special_tokens=True))
Read moreDetails
Benchmarks: refusals on 1/84 prompts, perplexity PPL on wikitext-2 is 152.8 (compared to 125.8 for the base).
Limitations: this is a 2B model, so it may make mistakes on simple facts; the visual and audio parts were not tested by the author.
Training details: the llm-abliteration tool was used with the recipe/gemma4-ple.patch patch to account for per_layer_projection.
Discuss Telegram 0 4

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 9.5 GB · ready files11 GB
  • FP8 / INT8recommendedweights 4.8 GB5.7 GB
  • 4-bitweights 2.7 GB3.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2080 Ti

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # wafflefrisbee/gemma-4-E2B-it-abliterated
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download wafflefrisbee/gemma-4-E2B-it-abliterated --local-dir ~/models/gemma-4-E2B-it-abliterated
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. gemma-4-E2Bfamily root
  2. gemma-4-E2B-it
  3. gemma-4-E2B-it-abliterated

New posts about versions of gemma-4-E2B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

wafflefrisbee/gemma-4-E2B-it-abliterated
Author
wafflefrisbee
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

40

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments