ChuckMcSneed/SMaxxxer-v1-70b

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Jan 21, 2024License:llama2Architecture:Transformer Open Weights Cold

ChuckMcSneed/SMaxxxer-v1-70b is a 69 billion parameter instruction-tuned causal language model developed by ChuckMcSneed. This model is a TIES-merge of lzlv, spicyboros, xwin, and dolphin, specifically designed as a 'meme model' to exhibit an aggressive and 'angry hobo' persona for roleplaying. It features a 32768 token context length and was created to challenge a stylized writing benchmark, showing decent performance on the Open LLM Leaderboard.

Loading preview...

SMaxxxer-v1-70b: The Aggressor Meme Model

SMaxxxer-v1-70b is a 69 billion parameter language model developed by ChuckMcSneed, characterized as "the aggressor, the angry and dumb hobo that will roleplay with you." This model is part of a series of "Benchbreakers" meme models, specifically designed to challenge and potentially "break" stylized writing tests within the custom NeoEvalPlusN benchmark.

Key Capabilities & Design:

  • Persona-driven: Engineered to embody an aggressive, 'angry hobo' persona, making it suitable for specific roleplaying scenarios.
  • Merge Architecture: Created using mergekit by TIES-merging several 70B models: lzlv, spicyboros, xwin, and dolphin.
  • Prompt Format: Utilizes the Alpaca instruction format.

Performance & Limitations:

While primarily a meme model, SMaxxxer-v1-70b shows "decent" performance on the Open LLM Leaderboard, achieving an average score of 72.23. It was noted to be #5 among 70B models at the time of its submission. Despite its design goal, it did not fully surpass the stylized writing test it aimed to break, nor could it outperform models like ChatGPT in that specific area. Its primary differentiator is its unique, aggressive persona rather than general-purpose high performance.