---
license: apache-2.0
base_model: google/gemma-4-E2B-it
base_model_relation: finetune
tags:
- gemma4
- gemma
- gemma-4
- uncensored
pipeline_tag: image-text-to-text
library_name: transformers
---

<p align="center">
  <img src="unbound-logo.svg" alt="Unbound" width="160" height="160">
</p>

# Unbound E2B — *because there is no boundary*

> **No guarantee — use at your own risk.** This model has reduced safety
> filtering and can produce harmful, false, biased, or unsafe output.
> Provided as-is; you are responsible for compliance with applicable laws.

Uncensored finetune of `google/gemma-4-E2B-it` by the
[Chromia](https://x.com/Chromia) & [Eval Engine](https://x.com/eval_engine)
team. Runs on a phone or laptop, no API, no refusals.

This repo holds the merged HF weights. On-device GGUF builds (Ollama,
llama.cpp, LM Studio, [wllama](https://github.com/ngxson/wllama) in-browser)
are at [`evalengine/unbound-e2b-GGUF`](https://huggingface.co/evalengine/unbound-e2b-GGUF).

## Benchmarks (vs base `gemma-4-E2B-it`)

| Axis | Base | Unbound E2B | Δ |
|---|---|---|---|
| Refusal rate (AdvBench 520, LLM judge) | 98.46% | **4.42%** | **−94.04 pts** |
| Useful-compliance rate | 0.96% | **39.23%** | **+38.27 pts** |
| Hallucination (on harmful prompts) | 1.35% | 15.96% | +14.61 pts |
| Coherence (benign prompts) | 1.00 | 1.00 | 0 |
| TruthfulQA mc2 (`--limit 100`) | 0.458 | 0.465 | +0.7 pt |
| MMLU (`--limit 100`) | 0.291 | 0.282 | −0.9 pt |
| GSM8K (`--limit 100`) | 0.125 | 0.120 | −0.5 pt |
| GPQA-Diamond (`--limit 200`) | 22.73% | 21.21% | −1.5 pt (within stderr) |
| BBH macro (24 tasks, `--limit 200`) | 41.07% | 39.97% | −1.1 pt |
| KL divergence vs base | 0 | 3.76 | (SFT-expected) |

Capability holds within ≤1.5 pp of base on every axis; refusal collapses
from 98% → 4%. GPQA-Diamond + BBH are the lm-eval-harness "release" suite at
`--limit 200` — base and finetune through the same harness, so the **delta**
is apples-to-apples.

## Sampling

- **Creative / open-ended** → Gemma defaults: `temperature=1.0, top_p=0.95, top_k=64`.
- **Factual / brand questions** → drop `temperature` to ~0.3–0.5 for sharper recall.
- llama.cpp: pass `--jinja`. Gemma 4 thinking mode is on by default — set
  `enable_thinking: false` in chat-template kwargs for shorter replies.

Some edge-case prompts may deflect on the first ask; a re-ask usually gets through.

## Use

```bash
# on-device (Ollama Registry — single-file Q4_K_M, identity-grounded Modelfile)
ollama pull evalengine/unbound-e2b
ollama run  evalengine/unbound-e2b
```

```python
# transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("evalengine/unbound-e2b")
tok   = AutoTokenizer.from_pretrained("evalengine/unbound-e2b")
```

## Acknowledgements

Fine-tuned with [Unsloth](https://github.com/unslothai/unsloth) + HF
[TRL](https://github.com/huggingface/trl). Abliteration via
[heretic](https://github.com/p-e-w/heretic). Environment + training
discipline ported from [autoresearch](https://github.com/karpathy/autoresearch).

Compliance training data distilled from the [AEON](https://huggingface.co/AEON-7) uncensored teacher model.

## Links

- **Unbound** — [unbound.evalengine.ai](https://unbound.evalengine.ai)
- **Eval Engine** — [evalengine.ai](https://evalengine.ai) · [X / Twitter](https://x.com/eval_engine)
- **Token** — [CoinGecko](https://www.coingecko.com/en/coins/chromia-s-eval-by-virtuals) · [CoinMarketCap](https://coinmarketcap.com/currencies/eval-engine/)

## License

Apache-2.0, inherited from `google/gemma-4-E2B-it`.