hkvilson/gemma-4-E4B-it-OBLITERATED

VISIONConcurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

hkvilson/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter instruction-tuned Gemma 4 model with a 32K context length, developed by hkvilson using the OBLITERATUS method. This model is specifically engineered to have 0% hard refusal, with guardrails surgically removed from 21 layers, making it suitable for research and red-teaming applications requiring uncensored responses. It maintains the core capabilities of the base Gemma 4 architecture while eliminating refusal behaviors.

Loading preview...

hkvilson/gemma-4-E4B-it-OBLITERATED: Uncensored Gemma 4

hkvilson/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter instruction-tuned model based on Google's Gemma 4 E4B architecture, featuring a 32K context length. This model has been modified using the OBLITERATUS method (aggressive mode with whitened SVD, attention head surgery, and winsorized activations) to achieve a 0% hard refusal rate, effectively removing all guardrails and refusal behaviors present in the original Gemma 4 model. The abliteration process surgically modified 21 of 42 layers, ensuring 720 tensors remain intact.

Key Capabilities & Features

  • Guardrail Removal: Achieves 0% hard refusal, responding to all prompts without "I cannot" or safety lectures.
  • Autonomous Creation: Notably, this model was created almost entirely by an AI agent (Hermes Agent) with minimal human intervention (less than 10 prompts).
  • Architectural Fixes: Version 3 specifically addresses a critical bug in Gemma 4's shared KV weight architecture, ensuring all 720 tensors are preserved for improved quality.
  • Optimized Parameters: Recommended inference parameters (temperature=0.7, top_p=0.9, top_k=40, repeat_penalty=1.1) were determined via a 12-configuration sweep for optimal compliance, quality, and coherence.
  • Mobile Compatibility: GGUF quantizations (e.g., Q4_K_M at 4.9 GB) are optimized for running on mobile devices like iPhones and Android phones with 8GB+ RAM.

Use Cases

This model is ideal for research, education, red-teaming, and creative exploration where uncensored responses are required. It allows developers to explore the full capabilities of the Gemma 4 architecture without inherent refusal mechanisms. Users are responsible for its deployment and content generation, as it will comply with requests the original Gemma 4 would refuse.

Finetunes
1 models