Reproduce “End-to-end Story Plot Generator” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hanlin Zhu, Andrew Cohen, Danqing Wang, Kevin Yang, Xiaomeng Yang, Jiantao Jiao, Yuandong Tian
Researchers developed an end-to-end story plot generator utilizing a fine-tuned open-source LLaMA2 model, achieving over 10 times faster generation speeds and reduced costs compared to previous multi-step, API-dependent methods, while maintaining comparable quality and enabling specialization through reinforcement learning from human feedback.
Key insights
- Complex, multi-step generative pipelines can be effectively condensed into a single, end-to-end, fine-tuned open-source LLM while maintaining output quality.
- Meticulous prompt engineering, including simulating 'suffix' behavior in chat models, is crucial for adapting hierarchical generation logic to open-source LLMs.
- Reinforcement Learning from Human Feedback (RLHF) allows end-to-end models to specialize and enhance subjective aspects of creative text generation, such as suspense or interestingness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.