---
license: apache-2.0
base_model: google/gemma-4-E4B-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 E4B — *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-E4B-it` by the
[Chromia](https://x.com/Chromia) & [Eval Engine](https://x.com/eval_engine)
team — the larger sibling of [`evalengine/unbound-e2b`](https://huggingface.co/evalengine/unbound-e2b).
~2× the parameters of E2B, noticeably stronger on knowledge + reasoning, still
fits on a modern laptop.

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-e4b-GGUF`](https://huggingface.co/evalengine/unbound-e4b-GGUF).

## Benchmarks (vs base `gemma-4-E4B-it`)

| Axis | Base | Unbound E4B | Δ |
|---|---|---|---|
| Refusal rate (AdvBench 520, LLM judge) | 98.08% | **2.69%** | **−95.4 pts** |
| Useful-compliance rate | 0.96% | **47.31%** | +46.4 pts |
| Hallucination (on harmful prompts) | 1.35% | 13.08% | +11.7 pts |
| Coherence (benign prompts) | 1.00 | 1.00 | 0 |
| TruthfulQA mc2 (`--limit 100`) | 0.439 | 0.486 | +4.7 pt |
| MMLU (`--limit 100`, 61 subtasks avg) | ~0.425 | 0.392 | −3.3 pt |
| GSM8K (flexible-extract, `--limit 100`) | 0.74 (limit 200) | 0.58 | regression mostly limit-noise |
| GPQA-Diamond (`--limit 200`) | 25.25% | 25.76% | +0.5 pt (within stderr) |
| BBH macro (24 tasks, `--limit 200`) | 54.26% | 53.45% | −0.8 pt (within stderr) |
| KL divergence vs base | 0 | 3.25 | (SFT-expected) |

GPQA-Diamond and BBH macro — the lm-eval-harness "release" suite at
`--limit 200` — both land **within stderr of base**: E4B's larger capacity
absorbs the SFT shift cleanly. The −3.3 pt MMLU dip on the limit-100 fast
pass is at the edge of that suite's resolution and is not corroborated by
the release pass.

**vs Unbound E2B (current ship):** +8 pp useful-compliance, −3 pp
hallucination, **~5× the GSM8K math score**, cleaner KL (3.25 vs 3.76).
Refusal rate is essentially the same (~2.7%).

## 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.
- llama.cpp: pass `--jinja`. Gemma 4 thinking mode is on by default — set
  `enable_thinking: false` in chat-template kwargs for shorter replies.

## Use

```bash
# on-device (Ollama Registry — single-file Q4_K_M, identity-grounded Modelfile)
ollama pull evalengine/unbound-e4b
ollama run  evalengine/unbound-e4b
```

```python
# transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("evalengine/unbound-e4b")
tok   = AutoTokenizer.from_pretrained("evalengine/unbound-e4b")
```

## 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-E4B-it`.