GitHub - aiwaves-cn/RecurrentGPT: Official Code for Paper: RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text
🌟 Dive into the world of long text generation with GitHub's RecurrentGPT! 📝🤖 Simulating LSTM RNN with natural language, this tool offers interactive generation of extensive text. Enhance your writing experience and unleash your creativity! #AI #textgeneration #RecurrentGPT
- RecurrentGPT simulates LSTM RNN with natural language, using paragraphs of text to replace vectorized elements and prompt engineering for recurrence mechanism.
- It operates by receiving paragraphs and a brief plan at each timestep, attending to long-term memory, and updating short-term memory for key information.
- RecurrentGPT generates new paragraphs, plans, and updates memories using a prompt for the backbone LLM.
- It offers flexibility in human-AI co-writing by generating initial paragraphs and allowing writers to edit or provide their own plans.
- Customizations are possible via script configurations like the number of rounds, output file, topic, and type of novel.
- The tool provides web demos and instructions for local deployment with customized LLMs.
- Citations and acknowledgments are included for the RecurrentGPT model.
- Various details, such as contributors, code resources, and licensing information, are available in the repository.