crestf411/MN-Slush

crestf411/MN-Slush is a 12 billion parameter, two-stage fine-tuned language model based on Mistral-Nemo-Base-2407, developed by crestf411. It is specifically optimized to enhance creativity, writing capabilities, and roleplaying performance through a unique LoRA dropout training methodology. The model leverages a continued pretraining stage to boost creative output, followed by a fine-tuning stage to refine instruction adherence and roleplaying, making it suitable for generative text applications requiring imaginative and interactive responses.