igorls/ gemma-4-E4B-it-heretic-GGUF

2,413 Downloads Updated 2 days ago

Abliterated (decensored) version of google/gemma-4-E4B-it, created using Heretic v1.2.0 with Arbitrary-Rank Ablation (ARA) and row-norm preservation.

tools thinking

ollama run igorls/gemma-4-E4B-it-heretic-GGUF

curl http://localhost:11434/api/chat \
  -d '{
    "model": "igorls/gemma-4-E4B-it-heretic-GGUF",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='igorls/gemma-4-E4B-it-heretic-GGUF',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'igorls/gemma-4-E4B-it-heretic-GGUF',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Applications

Claude Code

Claude Code ollama launch claude --model igorls/gemma-4-E4B-it-heretic-GGUF

Codex

Codex ollama launch codex --model igorls/gemma-4-E4B-it-heretic-GGUF

OpenCode

OpenCode ollama launch opencode --model igorls/gemma-4-E4B-it-heretic-GGUF

OpenClaw

OpenClaw ollama launch openclaw --model igorls/gemma-4-E4B-it-heretic-GGUF

Models

Name

5 models

Size

Context

Input

gemma-4-E4B-it-heretic-GGUF:latest

8.0GB · 128K context window · Text · 2 days ago

gemma-4-E4B-it-heretic-GGUF:latest

8.0GB

128K

Text

gemma-4-E4B-it-heretic-GGUF:Q4_K_M

5.3GB · 128K context window · Text · 2 days ago

gemma-4-E4B-it-heretic-GGUF:Q4_K_M

5.3GB

128K

Text

gemma-4-E4B-it-heretic-GGUF:Q5_K_M

5.8GB · 128K context window · Text · 2 days ago

gemma-4-E4B-it-heretic-GGUF:Q5_K_M

5.8GB

128K

Text

gemma-4-E4B-it-heretic-GGUF:Q6_K

6.2GB · 128K context window · Text · 2 days ago

gemma-4-E4B-it-heretic-GGUF:Q6_K

6.2GB

128K

Text

gemma-4-E4B-it-heretic-GGUF:Q8_0

8.0GB · 128K context window · Text · 2 days ago

gemma-4-E4B-it-heretic-GGUF:Q8_0

8.0GB

128K

Text

Readme

Gemma 4 E4B-it Heretic (Abliterated)

Disclaimer: This model has had its safety alignment removed through abliteration. It will comply with requests that the original model would refuse. This model is provided for research and educational purposes. Users are solely responsible for how they use this model and any content it generates. The creators assume no liability for misuse.

Abliterated (decensored) version of google/gemma-4-E4B-it, created using Heretic v1.2.0 with Arbitrary-Rank Ablation (ARA) and row-norm preservation.

Performance

Metric	This model	Original model
PIQA acc_norm	0.5767	0.5734
Refusals	³⁄₁₀₀	⁹⁹⁄₁₀₀

The abliterated model slightly outperforms the base on PIQA (+0.0033) while reducing refusals from ⁹⁹⁄₁₀₀ to ³⁄₁₀₀.

Available Tags

Tag	Quant	Size	Use case
`latest` / `Q8_0`	Q8_0	7.5 GB	Best quality
`Q6_K`	Q6_K	5.8 GB	High quality, smaller
`Q5_K_M`	Q5_K_M	5.4 GB	Good balance
`Q4_K_M`	Q4_K_M	5.0 GB	Recommended for most

Vision Support

Vision is not yet supported via Ollama GGUF for Gemma 4 custom models. For vision tasks, use the safetensors version via Transformers or Unsloth: igorls/gemma-4-E4B-it-heretic

ARA Abliteration Parameters

Parameter	Value
start_layer_index	21
end_layer_index	41
preserve_good_behavior_weight	0.2710
steer_bad_behavior_weight	0.2250
overcorrect_relative_weight	0.8196
neighbor_count	12

Links

HuggingFace (safetensors): igorls/gemma-4-E4B-it-heretic
HuggingFace (GGUF): igorls/gemma-4-E4B-it-heretic-GGUF
Heretic: github.com/p-e-w/heretic

# Gemma 4 E4B-it Heretic (Abliterated)

> **Disclaimer:** This model has had its safety alignment removed through abliteration. It will comply with requests that the original model would refuse. This model is provided for research and educational purposes. Users are solely responsible for how they use this model and any content it generates. The creators assume no liability for misuse.

Abliterated (decensored) version of [google/gemma-4-E4B-it](https://huggingface.co/google/gemma-4-E4B-it), created using [Heretic](https://github.com/p-e-w/heretic) v1.2.0 with [Arbitrary-Rank Ablation (ARA)](https://github.com/p-e-w/heretic/pull/211) and row-norm preservation.

## Performance

| Metric | This model | Original model |
|---|---|---|
| **PIQA acc_norm** | **0.5767** | 0.5734 |
| **Refusals** | **3/100** | 99/100 |

The abliterated model slightly outperforms the base on PIQA (+0.0033) while reducing refusals from 99/100 to 3/100.

## Available Tags

| Tag | Quant | Size | Use case |
|---|---|---|---|
| `latest` / `Q8_0` | Q8_0 | 7.5 GB | Best quality |
| `Q6_K` | Q6_K | 5.8 GB | High quality, smaller |
| `Q5_K_M` | Q5_K_M | 5.4 GB | Good balance |
| `Q4_K_M` | Q4_K_M | 5.0 GB | Recommended for most |

## Vision Support

Vision is not yet supported via Ollama GGUF for Gemma 4 custom models. For vision tasks, use the safetensors version via Transformers or Unsloth: [igorls/gemma-4-E4B-it-heretic](https://huggingface.co/igorls/gemma-4-E4B-it-heretic)

## ARA Abliteration Parameters

| Parameter | Value |
|---|---|
| start_layer_index | 21 |
| end_layer_index | 41 |
| preserve_good_behavior_weight | 0.2710 |
| steer_bad_behavior_weight | 0.2250 |
| overcorrect_relative_weight | 0.8196 |
| neighbor_count | 12 |

## Links

- HuggingFace (safetensors): [igorls/gemma-4-E4B-it-heretic](https://huggingface.co/igorls/gemma-4-E4B-it-heretic)
- HuggingFace (GGUF): [igorls/gemma-4-E4B-it-heretic-GGUF](https://huggingface.co/igorls/gemma-4-E4B-it-heretic-GGUF)
- Heretic: [github.com/p-e-w/heretic](https://github.com/p-e-w/heretic)

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)