Closest to this would be https://www.eleuther.ai whose training data is largely public and training processes are openly discussed, planned, and evaluated on their Discord server. Much of their training dataset is available at https://the-eye.eu (their onion link is considered "primary", however, due to copyright concerns)