Hugging Face undertook the reproduction of 2,200 research papers presented at the International Conference on Machine Learning (ICML). The project involved creating publicly available implementations and datasets for each paper. These implementations utilize various models and agent systems. The resulting codebases are available under an open-weights license. The project aims to facilitate independent verification and further research. The data and models are designed to be used for inference and experimentation.
Source: https://huggingface.co/blog/icml-2026-open-reproductions