SicariusSicariiStuff/Assistant_Pepe_70B
Assistant_Pepe_70B by SicariusSicariiStuff is a 70 billion parameter language model based on the Llama-3.1-70B architecture, uniquely fine-tuned to maximize helpfulness and 'shitposting' while reducing positivity. It demonstrates significant lateral thinking capabilities, outperforming other models on complex trick questions without explicit training data. This model excels in creative writing, bantering, and general tasks, offering exceptional coherency up to 64K context length.
Loading preview...
Overview
Assistant_Pepe_70B is a 70 billion parameter language model developed by SicariusSicariiStuff, built upon the Llama-3.1-70B base. It is specifically designed to combine high helpfulness with a 'shitposting' style, featuring a distinct negativity bias and minimal censorship (rated 9.5/10). The model has achieved a #1 ranking globally for 70B models on the UGI benchmark, even outperforming base Llama-3.3-70B-Instruct and Llama-3.1-70B-Instruct, and nearly matching Mistral-Large-Instruct-2411 in raw intelligence.
Key Capabilities
- Exceptional Lateral Thinking: Demonstrates a unique ability to solve complex trick questions that often stump other frontier models, even without specific training data for these problems.
- Engaging Interaction: Known for its exquisite bantering, creative writing, and ability to 'roast' users due to its negativity bias infusion.
- High Coherency: Leverages the Llama 3.1 70B base for excellent long context understanding, maintaining superb coherency at 32K and very good coherency at 64K.
- Code Generation: Capable of shipping code.
- No System Prompt Required: Simplifies interaction by not needing a system prompt.
Good For
- Shitposting and General Tasks: Ideal for users seeking a model with a unique, quirky, and unfiltered personality.
- Creative Writing: Excels in generating highly creative and unique text.
- Problem Solving: Particularly useful for tasks requiring unconventional or 'lateral' thinking.
- Uncensored Interactions: Suitable for use cases where minimal censorship is desired.