Open-Assistant

Why we included this project

OpenAssistant was one of the most visible community efforts to build an open alternative to commercial chat assistants. The repo holds the whole stack behind that project: a web frontend, an API, and the crowdsourced data-collection and RLHF training pipeline that shaped the assistant from human-written conversations. Anyone trying to understand how a human-feedback loop for fine-tuning a language model works end to end will find a concrete example here, not just a model card. The released oasst2 dataset, with millions of human-written prompts and ranked responses, is also worth pulling on its own for training or evaluating instruction-following models. Development has stopped, so treat it as a study reference and data source rather than something to build a live product on.