The collator function in collator.py will drop samples in train dataset, even validation/test dataset ?

Hi authors,
Thank you for your great work! I'm working on PCBA dataset with your v1 code. But I noticed that the collator function used to load samples for Batch in collator.py would drop samples in training dataset, even validation/test dataset. The code snapshot is as blow:
def collator(items, max_node=512, multi_hop_max_dist=20, spatial_pos_max=20): items = [ item for item in items if item is not None and item.x.size(0) <= max_node]
Since the parameter max_node was set as 128 for PCBA dataset, it meant samples with item.x.size(0) larger than 128 wouldn't be loaded. I did a data exploration and found there were some samples whose x.size(0) is larger than 128 in both training dataset and validation/test dataset. I think it might also affect the evaluation score on the leaderboard. Or do I misunderstand it?. Thank you!

Start with collator.py and the PCBA dataset configuration where max_node is set to 128. Inspect how samples are filtered for the training, validation, and test splits, then measure which samples are omitted. Done means establishing whether the filtering is intentional and documenting or correcting its effect on evaluation.