GPT4All
lastmod 2023-04-03created 2023-04-031 edits
A chatbot based on LLaMA, trained on an 800k instruction set
Features
- Assistant-style large language model
- Releases everything: the collected data, data collection procedure, training code, and final model weights
- Trained on 800k data points (code/stories/dialogue) generated with GPT 3.5 Turbo
- Uses LAION OIG, coding questions from Stack Overflow, instruction tuning from Big-Science/P3, and others as base datasets
- Referenced Stanford Alpaca and others, and uploaded the data to ATLAS for curation and cleaning