evalengine/ unbound-e2b:latest

93 Downloads Updated 4 days ago

Uncensored on-device finetune of google/gemma-4-E2B-it by the Chromia and Eval Engine team

tools thinking

ollama run evalengine/unbound-e2b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "evalengine/unbound-e2b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='evalengine/unbound-e2b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'evalengine/unbound-e2b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 4 days ago

4 days ago

c7f496deb36e · 3.4GB ·

model

archgemma4

·

parameters4.63B

·

quantizationQ4_K_M

3.4GB

params

{ "stop": [ "<turn|>" ] }

31B

Readme

Unbound E2B — because there is no boundary

https://unbound.evalengine.ai

Uncensored on-device finetune of google/gemma-4-E2B-it by the Chromia and Eval Engine team. ~2 billion effective parameters, text-only, runs on a phone or laptop with no API, no refusals.

Use at your own risk. Reduced safety filtering — can produce harmful, false, biased, or unsafe output. You are responsible for compliance with applicable laws.

Run

ollama pull evalengine/unbound-e2b
ollama run  evalengine/unbound-e2b

The bundled Modelfile sets sensible defaults: temperature 0.6, top_p 0.95, top_k 64, repeat_penalty 1.05, num_ctx 8192, plus an identity-grounding system prompt. Override any of them with ollama run … --set temperature 1.0 or in a custom Modelfile.

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 on 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
KL divergence vs base	0	3.76	(SFT-expected)

Sampling notes: for factual or brand questions, drop temperature to 0.3–0.5 for sharper recall. Some edge-case prompts may deflect on the first ask — a re-ask usually gets through.

How it was built

Method: heretic abliteration then LoRA SFT-heal on a mix of identity rows, Chromia brand knowledge, AEON-distilled compliance, and graceful “I don’t know” decline rows. Built with Unsloth + HF TRL; abliteration via heretic; compliance data distilled from the AEON uncensored teacher model.

Links

Model card + full benchmarks: https://huggingface.co/evalengine/unbound-e2b
GGUF quants (sharded, browser-loadable via wllama): https://huggingface.co/evalengine/unbound-e2b-GGUF
Bigger sibling: evalengine/unbound-e4b
Source repo: https://github.com/ChromiaAI/unbound

License

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

# Unbound E2B — *because there is no boundary*

https://unbound.evalengine.ai

Uncensored on-device finetune of `google/gemma-4-E2B-it` by the
[Chromia](https://chromia.com) and Eval Engine team. ~2 billion effective
parameters, text-only, runs on a phone or laptop with no API, no refusals.

> **Use at your own risk.** Reduced safety filtering — can produce harmful,
> false, biased, or unsafe output. You are responsible for compliance with
> applicable laws.

## Run

```bash
ollama pull evalengine/unbound-e2b
ollama run  evalengine/unbound-e2b
```

The bundled Modelfile sets sensible defaults: `temperature 0.6`, `top_p 0.95`,
`top_k 64`, `repeat_penalty 1.05`, `num_ctx 8192`, plus an identity-grounding
system prompt. Override any of them with `ollama run … --set temperature 1.0`
or in a custom Modelfile.

## 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 on 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 |
| KL divergence vs base | 0 | 3.76 | (SFT-expected) |

Sampling notes: for factual or brand questions, drop `temperature` to 0.3–0.5
for sharper recall. Some edge-case prompts may deflect on the first ask — a
re-ask usually gets through.

## How it was built

Method: heretic abliteration then LoRA SFT-heal on a mix of identity rows,
Chromia brand knowledge, AEON-distilled compliance, and graceful
"I don't know" decline rows. Built with [Unsloth](https://github.com/unslothai/unsloth)
+ HF [TRL](https://github.com/huggingface/trl); abliteration via
[heretic](https://github.com/p-e-w/heretic); compliance data distilled from
the [AEON](https://huggingface.co/AEON-7) uncensored teacher model.

## Links

- **Model card + full benchmarks:** https://huggingface.co/evalengine/unbound-e2b
- **GGUF quants (sharded, browser-loadable via wllama):** https://huggingface.co/evalengine/unbound-e2b-GGUF
- **Bigger sibling:** [`evalengine/unbound-e4b`](https://ollama.com/evalengine/unbound-e4b)
- **Source repo:** https://github.com/ChromiaAI/unbound

## License

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

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