DM-LLM-Tiny: A Specialized D&D Content Generator
DM-LLM-Tiny is a compact 1.1 billion parameter language model, built upon the TinyLlama-1.1B-Chat-v1.0 base, and uniquely fine-tuned for generating creative content for Dungeons & Dragons. Developed by JBHarris, this model leverages QLoRA (4-bit NF4 quantization with LoRA r=64) on a dataset of approximately 500 synthetic D&D instruction/response pairs generated by Claude.
Key Capabilities
This model is engineered to assist Dungeon Masters and players by generating a variety of D&D-specific content, including:
- NPCs: Crafting memorable non-player characters with backstories and motivations.
- Quests: Developing plot hooks, outlines, and full quest arcs.
- Dialogue: Creating in-character conversations, monologues, and banter.
- Locations: Describing vivid dungeons, towns, and wilderness settings.
- Encounters: Designing combat, social, and puzzle-based scenarios.
Good For
- Quick Idea Generation: Ideal for brainstorming sessions and generating initial concepts rapidly.
- D&D Enthusiasts: Provides a specialized tool for enhancing D&D campaigns.
- Resource-Constrained Environments: Its small size makes it suitable for deployment with limited hardware, such as via Ollama.
Limitations
As a 1.1 billion parameter model, DM-LLM-Tiny offers creative and fun outputs but does not match the quality or coherence of significantly larger models (7B+). It is best utilized as a supplementary tool for brainstorming rather than a complete replacement for human creativity and judgment.