Jommarn/UNSEEN_Gemma_4_E2B_NSFW

Model Overview

Jommarn/UNSEEN_Gemma_4_E2B_NSFW is a 5.1 billion parameter model derived from google/gemma-4-E2B-it, featuring a 32768 token context length. Its primary distinction is the complete and permanent removal of censorship and refusal mechanisms through a technique called Orthogonal Weight Projection (Abliteration). This modification was applied across all transformer layers (o_proj and down_proj), ensuring that the model's weights themselves are altered to prevent content refusal.

Key Capabilities

• Zero Censorship: Designed to generate explicit, NSFW, and taboo content without refusal, bypassing standard safety guardrails.
• vLLM Compatibility: The weight-level modification ensures full compatibility with vLLM and other optimized inference engines, as it does not rely on runtime hooks.
• Bilingual Support: Capable of generating highly explicit content in both English and Thai street slang.
• Vision-Language Integration: Retains its multimodal capabilities, allowing it to process images alongside text for descriptive generation.

Abliteration Methodology

The model's refusal direction was identified by analyzing mean activation differences between safe and NSFW prompts. This refusal vector was then mathematically subtracted from the output projections of all layers using an orthogonal projection formula. This process ensures the model is inherently incapable of representing refusal intent, leading to unfiltered outputs.

Intended Use

This model is intended for mature audiences and is designed for generating highly explicit and NSFW content. Users should exercise responsibility and ensure compliance with local laws and regulations. It is not intended for generating non-consensual imagery descriptions or illegal content.