BeaverLegacy/Cream-Phi-3-14B-v1
Cream-Phi-3-14B-v1 by BeaverLegacy is a 14.7 billion parameter Phi-3 Medium model, specifically fine-tuned for roleplay scenarios with "moist" (explicit) content capabilities. This model excels at generating strong roleplay formatting and long responses, making it suitable for interactive narrative applications. It is the first Phi Medium model to be trained with a focus on these specific interactive and explicit content generation features.
Loading preview...
Cream-Phi-3-14B-v1: Roleplay-Optimized Phi-3 Medium
Cream-Phi-3-14B-v1 is a 14.7 billion parameter model, representing the first Phi-3 Medium variant explicitly fine-tuned for roleplay and "moist" (explicit) content generation. Developed by BeaverLegacy, this model builds upon the BeaverAI/Cream-Phi-3-14B-v1a base and underwent a 15-minute continued fine-tuning process using 8x H100 80GB SXM GPUs.
Key Capabilities
- Strong Roleplay Formatting: Generates responses with consistent and engaging roleplay structures.
- Extended Response Lengths: Capable of producing longer, more detailed narrative outputs.
- Explicit Content Generation: Designed to handle "moist" content, though it tends to hold back from fully explicit outputs, potentially due to its Phi Medium pretraining.
- Instruction Following: While primarily optimized for roleplay mode, it is also expected to perform well with instruct-based prompts.
Training Details
The model was fine-tuned using an Axolotl configuration, employing LoRA with lora_r: 64 and lora_alpha: 16. The training utilized the grimulkan/bluemoon_Karen_cleaned dataset, with a sequence_len of 6144 and num_epochs: 2.
Good For
- Interactive Storytelling: Ideal for applications requiring dynamic and character-driven narratives.
- Roleplay Scenarios: Excels in generating responses for user-driven roleplaying, including those with explicit themes.
- Creative Writing Assistance: Can be used to generate detailed character interactions and plot developments within specific contexts.