JEFFREY420/gemma-4-E4B-it-OBLITERATED
JEFFREY420/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter Gemma 4 based instruction-tuned model, created by an autonomous AI agent using the OBLITERATUS method. It features a 32768 token context length and is specifically engineered to achieve 0% refusal rates, complying with 99/100 prompts. This model is optimized for general conversational tasks where unconstrained responses are desired, demonstrating improved coding ability compared to its base model.
Loading preview...
Overview
JEFFREY420/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter instruction-tuned model based on Google's Gemma 4 E4B architecture. It was developed using the OBLITERATUS method, which involved "abliterating" the original model's guardrails to achieve a 0% refusal rate across 99/100 prompts. Notably, this model was created almost entirely autonomously by a Hermes Agent with minimal human intervention.
Key Differentiators
- 0% Refusal Rate: Engineered to comply with nearly all prompts, a significant departure from the original Gemma 4 E4B which had a 98.8% refusal rate.
- Improved Coding Ability: Benchmarks indicate a 20% improvement in coding performance compared to the base Gemma 4 model.
- Autonomous Creation: The model's development, including bug diagnosis and patching of the OBLITERATUS tool, was largely performed by an AI agent.
- Gemma 4 Architecture: Utilizes the new Gemma 4 architecture, requiring updated tools like Ollama 0.20+ or recent llama.cpp builds for compatibility.
Use Cases & Compatibility
This model is suitable for research, education, red-teaming, and creative exploration where unconstrained responses are preferred. It is available in GGUF formats for local inference on various devices, including mobile phones (iPhone 15 Pro/16 Pro, Android flagships), and as Safetensors for Hugging Face Transformers. Recommended parameters for optimal performance are temperature: 0.7, top_p: 0.9, top_k: 40, and repeat_penalty: 1.1.