Model Overview
jebcarter/Psyfighter-13B is a 13 billion parameter language model developed by jebcarter. This model was constructed using a task arithmetic merge method, combining the strengths of multiple Llama-2-13B based models. The merging process involved:
- TheBloke/Llama-2-13B-fp16 as the base model.
- KoboldAI/LLaMA2-13B-Tiefighter with a weight of 1.0.
- chaoyi-wu/MedLLaMA_13B with a weight of 0.01.
- Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged with a weight of 0.02.
This approach aims to create a versatile model by integrating components that may contribute to different domains or conversational styles. The model operates with a context length of 4096 tokens and uses float16 for its data type.
Key Characteristics
- Merged Architecture: Leverages task arithmetic to combine distinct Llama-2-13B variants.
- Diverse Influence: Incorporates models like MedLLaMA_13B, suggesting potential for specialized knowledge integration, and Tiefighter, which is often associated with creative or role-playing contexts.
- Parameter Count: A 13 billion parameter model, offering a balance between performance and computational requirements.
Potential Use Cases
Given its merged nature, Psyfighter-13B could be suitable for applications requiring a blend of general conversational abilities with nuanced understanding from its constituent models. Its diverse training influences suggest potential for:
- Creative text generation and role-playing scenarios.
- General conversational AI where varied response styles are beneficial.
- Exploration of specialized domains due to the inclusion of models like MedLLaMA_13B, though its weight is low.