Overview
DS-Archive/pygmalion-2-supercot-limarpv3-gradient-13b is a 13 billion parameter model built upon the Llama 2 architecture. It is a gradient merge of three distinct models: PygmalionAI/pygmalion-2-13b, Doctor-Shotgun/llama-2-supercot-lora, and lemonilia/LimaRP-Llama2-13B-v3-EXPERIMENT. The merge was performed using Zaraki's zarakitools, with SuperCoT integrated into deeper layers and LimaRPv3 into shallower layers, each contributing an average weight of 0.5.
Key Capabilities
- Enhanced Roleplaying: Combines the strengths of Pygmalion 2 for character interaction with SuperCoT's reasoning and LimaRPv3's length control.
- Controllable Response Lengths: Users can specify desired response lengths (e.g., tiny, short, medium, long, huge) using a modifier in the prompt, offering fine-grained control over output verbosity.
- Flexible Prompt Formats: Supports multiple roleplaying prompt formats, including Alpaca instruction format (from LimaRP v3) and Pygmalion/Metharme format, as well as a system prompt-only approach.
- Stylistic Elements: Incorporates additional stylistic elements from the merged models, aiming for more nuanced and engaging character responses.
Good For
- Interactive Storytelling: Ideal for applications requiring dynamic and character-driven narratives.
- Roleplaying Chatbots: Excels in scenarios where the AI needs to maintain a consistent persona and engage in extended conversational roleplay.
- Creative Content Generation: Suitable for generating detailed and descriptive text within a defined character context.
Limitations
- The model may exhibit biases similar to those found in niche online roleplaying communities. It is not intended for factual information retrieval or providing advice.