LAION-AI/Open-Assistant

Built on the InstructGPT training pipeline, the project uses crowdsourced data collection—with community ranking and labeling of prompts and responses—to train instruction-following models that can handle complex tasks, tool integration, and dynamic information retrieval. The platform includes a web-based data collection interface, inference system, and REST API backend, with the final OASST2 dataset released on HuggingFace for reproducible model training. Now archived, it served as a collaborative research effort to create open, efficient LLM-based assistants capable of running on consumer hardware.

37,440 stars. No commits in the last 6 months.