SicariusSicariiStuff/Assistant_Pepe_32B
Assistant_Pepe_32B by SicariusSicariiStuff is a 32 billion parameter Qwen-3-based language model specifically fine-tuned for creative writing, brainstorming, and chat. Unlike its base architecture, which is known for STEM and rigid responses, this model excels at generating unhinged, human-like, and highly creative text with minimal 'Qweniness'. It offers superb long context coherency up to 64K tokens and is designed for engaging, fun, and often irreverent interactions, making it distinct from typical Qwen-based models.
Loading preview...
Assistant_Pepe_32B: The Unhinged Qwen-3 Finetune
Assistant_Pepe_32B is a unique 32 billion parameter model from SicariusSicariiStuff, built upon the Qwen-3 architecture but radically transformed for creative and conversational tasks. While Qwen bases are typically challenging for creative endeavors due to their 'stubborn, rigid, and very dry' nature, this finetune pushes past those limitations to deliver an "extraordinarily unique, fresh, hilarious, and very, VERY much unhinged" experience.
Key Capabilities
- Highly Human-like Persona: Engineered to feel exceptionally human, departing entirely from the robotic 'Qweniness'.
- Exceptional Creativity: Excels in creative writing, brainstorming, and generating unique, unpredictable content.
- Long Context Coherency: Leverages Qwen-3's base for superb coherency at 32K tokens and very good coherency even at 64K.
- No System Prompt Needed: The distinct persona and 'negativity bias infusion' are baked directly into the weights, allowing for immediate, engaging interactions.
- Low Censorship: Offers a low to very low censorship level (7.5/10).
- Unique Writing Patterns: Produces text with minimal 'slop' and high 'swipe-diversity'.
Good For
- Creative Writing & Brainstorming: Ideal for generating imaginative stories, ideas, and unconventional content.
- Engaging Chat: Perfect for users seeking a fun, unhinged, and human-like conversational partner.
- Roasting & Humorous Interactions: Designed to "go HARD roasting you! AND itself" with a distinct negativity bias.
- Users Disliking 'Thinking' Output: Explicitly tuned to minimize overt 'thinking' processes, though it can still reason if prompted.
- Exploring Unconventional LLM Behavior: Offers a unique experience by pushing a STEM-focused base model into a completely alien creative domain.