Model Overview
chargoddard/storytime-13b is a 13 billion parameter chat model primarily focused on storytelling. It is constructed from a unique blend of several base models and LoRAs, specifically designed to enhance its narrative generation capabilities.
Key Components & Enhancements
This model's architecture is a composite, starting with the Chronorctypus-Limarobormes-13b base. It incorporates a significant contribution from ReMM-v2.2-L2-13B via SLERPing, and integrates a Llama-2-13B-Storywriter-LORA at a 0.5 weighting, alongside an ongoing work-in-progress storytelling LoRA. This combination aims to provide a robust foundation for creative text generation.
Performance Metrics
Evaluated on the Open LLM Leaderboard, chargoddard/storytime-13b demonstrates a balanced performance across various benchmarks, with an average score of 50.55. Notable scores include:
- ARC (25-shot): 62.03
- HellaSwag (10-shot): 83.96
- MMLU (5-shot): 57.48
- TruthfulQA (0-shot): 52.5
- Winogrande (5-shot): 75.53
While its GSM8K (8.34) and DROP (14.0) scores indicate areas for improvement in complex reasoning and reading comprehension, its overall profile suggests a strong aptitude for general language tasks, particularly those requiring creative output.
Intended Use Cases
This model is best suited for applications requiring narrative generation, creative writing, and interactive storytelling. Its fine-tuning for storytelling makes it a strong candidate for chatbots designed for role-playing, generating fictional content, or assisting with creative writing prompts. It is optimized to respond well to the Alpaca prompt format, facilitating ease of integration into existing workflows that utilize this structure.