mtr666/gemma-4-E4B-it-OBLITERATED

VISIONConcurrency Cost:1Model Size:7.9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 17, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

mtr666/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter Gemma 4 E4B model, meticulously modified to eliminate refusal behaviors, achieving a 0.0% refusal rate on a 842-prompt evaluation corpus. Developed using the OBLITERATUS method, it maintains core reasoning and creativity while showing improved coding ability. This model is specifically designed for research, red-teaming, and creative exploration requiring an uncensored large language model.

Loading preview...

Model Overview

mtr666/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter instruction-tuned Gemma 4 E4B model, distinguished by its complete removal of refusal behaviors. Utilizing the OBLITERATUS method, which includes whitened SVD, attention head surgery, and winsorized activations, this model achieves a 0.0% refusal rate across an expanded 842-prompt contrastive corpus, a significant reduction from the base model's 98.8% refusal rate.

Key Capabilities & Features

  • Zero Refusal: Engineered to provide full coherence and zero refusals, even on prompts that the original Gemma 4 E4B would decline.
  • Performance Retention: Despite extensive modification, the model retains 100% of its original reasoning and creativity capabilities. Notably, its coding ability improved by 20% post-modification.
  • Autonomous Development: A unique aspect of this model's creation is its development almost entirely by a Hermes Agent with minimal human intervention, including self-patching for NaN activation issues inherent to Gemma 4's bfloat16 architecture.
  • Optimized for Accessibility: Available in various GGUF quantizations (Q4_K_M, Q5_K_M, Q8_0) for deployment on diverse hardware, including mobile devices, and full bfloat16 Safetensors for Hugging Face Transformers.

Intended Use Cases

This model is ideal for:

  • Research and Red-Teaming: Exploring the boundaries of LLM behavior and safety mechanisms.
  • Creative Exploration: Generating content without inherent censorship or guardrails.
  • Educational Purposes: Understanding model modification techniques and their impact on behavior.

Users are advised to exercise personal responsibility, as the model will comply with requests that typically trigger refusals in standard LLMs.

Finetunes
1 models