More experimental configs

[qiangbo1222]

Hi,
Thanks a lot for providing the code for your paper. However, we found it hard to reproduce the results in your paper.
Could you provide the exact random seed for the experiments in Table 2 and Table 5? Besides, the code for training OT-CFM on the sampled data points to get the NLL metric is also missing. It would be nice of you to provide the code for this evaluation.

Best
Bo

[YuX-Ren]

Same question here. Especially, it's hard to reproduce the LJ experiments results. The training process is extremely unstable.

[RishalAggarwal]

Same here, awesome work! But it would be nice if you could provide the config/code used to train the OT-CFM model on sampled data points as well

[jarridrb]

Hi everyone, thanks for raising these concerns. We found there were a couple of issues with the codebase for the LJ task (the code base is an adaptation of a bunch of notebooks used for the paper) and also realize it wasn't clear how to run CFM. We just merged an update to the codebase fixing the bugs with LJ55 and also provided more documentation regarding running CFM. Let us know if there are other issues!

[jarridrb]

I'm closing this issue due to inactivity, but if anyone has further questions on this please feel free to reopen it!