KaraKaraWarehouse/MythaKiCOTlion-v2

MythaKiCOTlion-v2: A Merged 13B Language Model

MythaKiCOTlion-v2 is a 13 billion parameter decoder-only language model developed by KaraKaraWitch. It is a unique LoRA merge combining the Mythalion 13B base model with the SuperCOT LoRA by kaiokendev and the Kimiko v2 LoRA by nRuaif. This merge aims to integrate the distinct capabilities of its components into a single, versatile model.

Key Capabilities

• Instruction Following: Supports both Metharme and Alpaca instruction formats, allowing for flexible interaction and prompt engineering.
  • Metharme format: <|system|>Your system prompt goes here.<|user|>Are you alive?<|model|>
  • Alpaca format: ### Instruction:\nYour instruction or question here.\n### Response:
• Merged Architecture: Benefits from the combined characteristics of Mythalion 13B, known for its generative abilities, and the specialized enhancements provided by the SuperCOT and Kimiko v2 LoRAs.

Potential Use Cases

This model is suitable for developers and researchers looking for a 13B parameter model that combines different fine-tuning approaches. Its support for common instruction formats makes it adaptable for:

• General-purpose text generation
• Conversational AI applications
• Instruction-based tasks

Limitations and Considerations

As a merged model, its performance and specific biases are inherited from its constituent parts. Users should refer to the individual model cards for Mythalion 13B, SuperCOT LoRA, and Kimiko v2 LoRA for detailed information regarding their licenses, training, and potential limitations. The developer notes that its behavior is "untested territory," suggesting that performance may vary depending on the specific use case.