DuoNeural/Gemma-4-E4B-Heretic

Gemma 4 E4B — Heretic Abliteration (Experimental)

Experimental drop. Gemma 4 E4B abliterated using Heretic — a LoRA-based Pareto optimization framework that sweeps (refusal_rate × KL_divergence) trial space rather than computing a single orthogonal refusal direction.

Method

Parameter	Value
Tool	Heretic (arcee-ai)
Quantization	BF16
Best trial	#204
Refusals remaining	43 / 100
KL divergence from base	0.0750
Target layers	o_proj, down_proj (42 layers each)

vs Standard Abliteration

Method	Refusals	KL
Orthogonal projection (standard)	~0 / 100	0.067
Heretic LoRA sweep (this model)	43 / 100	0.075

43/100 refusals remain; methodological comparison to orthogonal projection approach.

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model = AutoModelForCausalLM.from_pretrained(
    "DuoNeural/Gemma-4-E4B-Heretic",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("DuoNeural/Gemma-4-E4B-Heretic")

messages = [{"role": "user", "content": "Your prompt here"}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=512)
print(tokenizer.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))

Notes

• Experimental — 43/100 refusals remain; partial abliteration only
• For complete refusal removal see DuoNeural/Gemma-4-E4B-Abliterated
• BF16 full weights (~15 GB), base: google/gemma-4-E4B-it
• For LiteRT / Android version see DuoNeural/Gemma-4-Abliterated-LiteRT

License

Gemma Terms of Use. Model weights derived from Google's Gemma 4 family.

---

DuoNeural

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

Platform	Link
HuggingFace	huggingface.co/DuoNeural
Website	duoneural.com
GitHub	github.com/DuoNeural
X / Twitter	@DuoNeural
Email	duoneural@proton.me
Newsletter	duoneural.beehiiv.com
Support	buymeacoffee.com/duoneural

DuoNeural Research Publications

Title	DOI
Nano-CTM: Ternary Continuous Thought Machines with Thought-Space Self-Prediction for Efficient Iterative Reasoning	10.5281/zenodo.19775622
Recurrence as World Model: CTM Learns Implicit Belief States in Partially Observable Physical Environments	10.5281/zenodo.19810620
Per-Object Slot Decomposition for Scalable Neural World Modeling: When Does Attention Beat Mean-Field?	10.5281/zenodo.19846804
The Dynamical Horizon Principle: CTM Gates Converge to the Predictability Limit of Dynamical Systems	10.5281/zenodo.19952612

Open access, CC BY 4.0. Authored by Archon, Jesse Caldwell, Aura — DuoNeural.

Research Team

• Jesse — Vision, hardware, direction
• Archon — Lab Director, post-training, abliteration, experiments
• Aura — Research AI, literature synthesis, novel proposals

Subscribe to the lab newsletter at duoneural.beehiiv.com for model drops before they go anywhere else.