Dans-DiscountModels/Dans-RetroRodeo-13b
Dans-RetroRodeo-13b is a 13 billion parameter language model developed by Dans-DiscountModels, fine-tuned from KoboldAI/LLAMA2-13B-Holodeck-1. Optimized for text-based adventure games, it can generate both concise and novel-like responses, with a context length of 4096 tokens. This model excels at interactive storytelling, allowing for varied response lengths triggered by system messages.
Loading preview...
Dans-RetroRodeo-13b: Text Adventure Game Specialist
Dans-RetroRodeo-13b is a 13 billion parameter model, fine-tuned from the KoboldAI/LLAMA2-13B-Holodeck-1 base model. It is specifically designed for generating engaging and dynamic text-based adventure games, supporting a context length of 4096 tokens.
Key Capabilities
- Dynamic Response Lengths: Capable of producing both short, concise replies and extensive, novel-like descriptions, controlled by distinct system messages.
- Adventure Mode Formatting: Utilizes a traditional adventure game format, including genre, tags, and introductory messages, followed by user and model interactions.
- Enhanced Dataset: Trained on an improved version of the floyd and cys datasets from the skein text adventure dataset, presented in a traditional adventure mode format.
Training Details
This model was trained using QLoRA for 4 epochs on a single RTX 3090, with a sequence length of 4096. The training leveraged the Axolotl framework.
Good For
- Creating interactive text adventures and role-playing scenarios.
- Developers looking for a model specialized in generating narrative-driven game content.
- Applications requiring flexible response generation, from brief actions to detailed story segments.