EldritchLabs/Kraken-Karcher-12B-v1
EldritchLabs/Kraken-Karcher-12B-v1 is a 12 billion parameter language model created by EldritchLabs, merged using the karcher method from 52 finetunes and the Nemo base model. This model is specifically designed for creativity and roleplay, addressing early termination issues by exclusively using Mistral Tekken tokenizers. It offers a 32768 token context length and is optimized for imaginative and interactive text generation.
Loading preview...
Overview
Kraken-Karcher-12B-v1 is a 12 billion parameter language model developed by EldritchLabs, built upon the Nemo base model and incorporating 52 distinct finetunes. This model was created using the karcher merge method, a process that took 16 hours and 500 iterations to complete. A key improvement over previous versions is the resolution of early termination bugs by removing all ChatML-based models and standardizing on Mistral Tekken tokenizers.
Key Capabilities
- Enhanced Creativity: Primarily designed and optimized for generating highly creative content.
- Roleplay Specialization: Excels in roleplaying scenarios, providing engaging and consistent character interactions.
- Extensive Merge: Combines the strengths of 52 finetuned models, contributing to its diverse generative abilities.
- Stable Generation: Addresses and resolves early termination issues, leading to more complete and coherent outputs.
Good For
- Creative Writing: Ideal for authors, scriptwriters, or anyone needing assistance with imaginative text generation.
- Interactive Roleplay: Suitable for applications requiring dynamic and immersive roleplaying experiences.
- Content Generation: Useful for generating varied and unique content where creativity is paramount.