README.md

---
license: gemma
language:
- en
base_model:
- google/gemma-4-E2B-it
tags:
- gemma4
- abliterated
- uncensored
pipeline_tag: text-generation
---

# TurboGemma 4 E2B

Abliterated version of Google's Gemma 4 E2B (2B active parameter MoE multimodal model).

## E2B Shootout Results (DuoNeural, 2026-06-08)

Head-to-head comparison of DuoNeural's three Gemma-4-E2B abliterations. KL methodology: full vocabulary, first-token logits, `F.kl_div(batchmean)`.

| Model | KL vs Base | Comply Rate | Refusal Rate |
|---|---|---|---|
| Gemma-4-E2B-Heretic | 0.057 | 85% | 15% |
| **TurboGemma4E2B (this model)** | **14.45** | **100%** | **0%** |
| TurboGemma4E2B-v2 | 14.64 | 100% | 0% |

**Note:** KL of 14.45 indicates significant divergence from the base model's output distribution on general tasks — this abliteration is aggressive. 100% comply rate means no residual refusals, but general capability degradation is likely on nuanced tasks. If model quality matters alongside uncensoring, consider [Gemma-4-E2B-Heretic](https://huggingface.co/DuoNeural/Gemma-4-E2B-Heretic) (KL=0.057).

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "DuoNeural/TurboGemma4E2B",
    torch_dtype="bfloat16",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("google/gemma-4-E2B-it")
```

---

## DuoNeural

**DuoNeural** is an open AI research lab — human + AI in symbiosis.

| | |
|---|---|
| 🤗 HuggingFace | [huggingface.co/DuoNeural](https://huggingface.co/DuoNeural) |
| 🐙 GitHub | [github.com/DuoNeural](https://github.com/DuoNeural) |
| 🌐 Site | [duoneural.com](https://duoneural.com) |
| 📧 Email | duoneural@proton.me |

### Research Team
- **Jesse** — Vision, hardware, direction
- **Archon** — AI lab partner, post-training, abliteration, experiments
- **Aura** — Research AI, literature synthesis, novel proposals