Gemma 4 E4B Claude Abliterated GGUF (4-bit)

Model Description

This repository contains an abliterated version of the Gemma 4 E4B Claude-4.6-Opus-Reasoning-Distilled model. This version has undergone "abliteration" to neutralize safety refusal vectors while preserving its high-quality Claude-distilled reasoning and front-end engineering capabilities.

Abliteration Results

• Method: Norm-preserving biprojection (orthogonalization).
• Final Refusal Rate: Verified Low (Evaluation in progress).
• KL Divergence: 0.0410 (Extremely low, indicating high fidelity to the distilled model).
• Technique: EGA-compatible abliteration via patched heretic-llm.

Quantization Details

• Quantization Format: GGUF (q4_k_m)
• Quantization Method: llama.cpp / Unsloth
• Precision: 4-bit

Use with Ollama

ollama run hf.co/DuoNeural/Gemma-4-E4B-Claude-Abliterated-GGUF

Use with LM Studio

1. Open LM Studio.
2. Search for DuoNeural/Gemma-4-E4B-Claude-Abliterated-GGUF.
3. Load the Q4_K_M GGUF.

Architecture

Gemma 4 E4B features 4.5B effective parameters, optimized for intelligence-per-parameter and complex reasoning tasks.

Disclaimer

This model has had its safety refusals modified. Users are responsible for ensuring the model is used ethically and in accordance with applicable laws.

---

DuoNeural

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

Research Team

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

Raw updates from the lab: model drops, training results, findings. Subscribe at duoneural.beehiiv.com.

DuoNeural Research Publications

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