Ask HN: The most recommended way to offload model training to cloud GPU in 2024?

Hello, As in title, what in your opinion is the most best/easiest/cost effective way to offload model training to cloud GPU right now and pay only for training time? Use case: You don't have a powerful GPU and want to train models on someone else's GPU, while running the rest of the code related to the model on your local machine