Delta-Vector/Hamanasu-QwQ-V2-RP
Hamanasu 32B by Delta-Vector is a chat-tuned language model, based on Delta-Vector/Hamanasu-QwQ-V1.5-Instruct, specifically fine-tuned for highly unconventional and 'brainrotted' chat interactions. It was trained on 10 million tokens, including data from platforms like Bsky, 4chan, and Discord logs, making it distinctively suited for humorous and non-standard conversational engagement rather than traditional roleplay.
Loading preview...
π Overview
Delta-Vector's Hamanasu 32B is a unique chat-tuned model, described as a "brainrotted" version of its base, Delta-Vector/Hamanasu-QwQ-V1.5-Instruct. It was fine-tuned using a diverse and unconventional dataset, including logs from Bsky, 4chan, and Discord, totaling 10 million tokens over 4 epochs on 8x H100 GPUs. This specialized training has resulted in a model optimized for highly idiosyncratic and humorous chat interactions.
Key Capabilities
- Unconventional Chat Partner: Designed to be a "highly dumb chat partner," excelling in non-standard and humorous conversations.
- ChatML Formatting: Utilizes ChatML for prompting, with a recommended system prompt for uncensored AI behavior.
- Optimized Sampler Preset: Recommends specific sampler settings (temperature: 1.8, min_p: 0.1) for optimal chat experience, suggesting a blank system prompt.
Good For
- Experimental Chat Applications: Ideal for use cases requiring a model with a distinct, often absurd, conversational style.
- Humorous Interactions: Suited for generating responses that are unexpected, funny, or unconventional, moving away from typical polite AI behavior.
- Non-Traditional Roleplay: While not for regular roleplay, its unique training makes it suitable for highly specific, unconventional interactive scenarios.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.