The issue regarding training time

[runfuYu]

Dear SeonghwanSeo:

May I ask what the number and model of Gpus you used for training your model are? How long does it take to train by calling the train_pocket_conditional file?

[SeonghwanSeo]

Sorry for delayed response. I didn't get any notifications...

The current model only supports a single GPU training. In fact, most computational cost arises from the reward function than RxnFlow model. In the paper, I used RTX 3080ti (12gb), 3090 (24gb), and a4000 (16gb). For target-specific tasks, I primarily used 3080ti and a4000. For pocket-conditional tasks, I remember that the training requires 2-4 days in RTX 3090, while the loss is converged much earlier. (I did not consider early stopping)