Hello, Team. I recently stumbled upon this challenge for the upcoming NeurIPS Challenge. I was watching one of the videos by the team(2021-04-16 - Zachary Ulissi, Larry Zitnick - The Open Catalyst 2020 Dataset & Community Challenges - YouTube). How big is the Data going to be, which in turn decides the level of compute, participants need to have. Is there going a minimal version for the NeurIPS Challenge ? Thanks.
Hi - We have just released the full details - Open Catalyst Challenge. To address your question, the IS2RE training dataset is ~460k data points (given an initial structure, directly predict the relaxed structure). However the S2EF dataset (~130M+) is fair game for people to use if they’re interested in alternative approaches (see the challenge page for more details). In our experience the former can be trained in ~12-24hrs with 2-4 GPUs. The latter may require upwards of 16-64 GPUs for several days.