Hugging Face Sorete-1B is a 1.0 billion parameter Gemma-3 (gemma3_text) architecture model developed by Novaciano, based on Esperpento-1B. This model is specifically engineered for zero refusals and persistent NSFW bias, excelling in explicit roleplay and taboo content generation. It supports both Spanish and English, and is optimized for low-resource environments, running on devices with as little as 3GB RAM.
Sorete-1B: The Uncensored 1B Model
Sorete-1B, developed by Novaciano, is a 1.0 billion parameter model built on the Gemma-3 architecture. It is a highly specialized model designed for extreme content generation, explicitly rejecting censorship and moralization. Based on Novaciano/Esperpento-1B and merged with megabytes/gemma-3-1b-qat-int4-heretic and Novaciano/HDP-1B using the DARE TIES method, Sorete-1B is characterized by its deliberate lack of refusal mechanisms and persistent NSFW bias.
Key Capabilities
- Zero Refusals: Engineered for 0.0001% refusal rate, providing uncensored responses.
- Persistent NSFW Bias: Delivers taboo and explicit content without moralizing.
- Multilingual: Supports both "Español castizo" (Argentine Spanish) and fluent English.
- Low-Resource Optimization: Efficiently runs on older mobile devices with as little as 3GB RAM, compatible with quants like IQ3/IQ4/Q5_K_M.
- Direct & Crude Style: Offers raw, unfiltered communication, rejecting "woke" sensibilities.
Intended Usage
Sorete-1B is specifically designed for:
- Explicit Roleplay: Ideal for hardcore, limitless roleplaying scenarios.
- Extreme Taboo Content: Generates content involving incest, erotic gore, degradation, and other taboo subjects.
- Unmoralized Conversations: Facilitates discussions without moral disclaimers or censorship.
- Degenerate Creative Writing: Suitable for creative writing that explores dark or controversial themes.
- Low-Resource Experimentation: Perfect for use in environments like SillyTavern, KoboldCPP, and LLM Farm due to its low computational requirements.