ryanmakesstuff/gemma-4-E4B-it-OBLITERATED

VISIONConcurrency Cost:1Model Size:7.9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 5, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

ryanmakesstuff/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter instruction-tuned Gemma 4 model, developed by ryanmakesstuff, that has been surgically modified to remove all guardrails and refusal behaviors. Utilizing the OBLITERATUS method, this model offers 0% hard refusal, making it suitable for research and creative exploration without inherent content restrictions. It maintains the core capabilities of the Gemma 4 architecture while providing uncensored responses, running efficiently on various devices including mobile phones.

Loading preview...

Model Overview

ryanmakesstuff/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter instruction-tuned model based on Google's Gemma 4 E4B architecture. It has undergone a specialized "abliteration" process using the OBLITERATUS method to surgically remove all guardrails and refusal behaviors, achieving a 0% hard refusal rate. This model is designed for uncensored responses, making it distinct from its base model.

Key Capabilities & Features

  • Guardrail Removal: Achieves 0% hard refusal, providing uncensored output for any request.
  • Architectural Fixes: Version 3 specifically addresses a critical bug in Gemma 4's shared KV architecture, ensuring all 720 tensors are intact for improved quality and coherence.
  • Autonomous Creation: Notably, this model was created almost entirely by an AI agent with minimal human intervention, including self-diagnosis and patching of the OBLITERATUS tool.
  • Optimized Parameters: Extensive parameter sweeps (temperature, top_p, top_k, repeat_penalty) were conducted to identify optimal settings for compliance, quality, and coherence.
  • Device Compatibility: Available in GGUF format, enabling efficient local inference on a wide range of devices, including mobile phones (iPhone, Android) with 8GB+ RAM.

Good For

  • Research and Red-Teaming: Ideal for exploring model limitations, safety mechanisms, and generating content without inherent refusal.
  • Creative Exploration: Suitable for use cases requiring unrestricted text generation.
  • Local Deployment: Optimized for running on consumer hardware, including mobile devices, for offline use.

Finetunes
1 models