The scheduler in the original codebase is called every batch: this means that the milestones are effectively 32x smaller. https://github.com/JiaxuanYou/graph-generation/blob/master/train.py#L218
The scheduler in the original codebase is called every batch: this means that the milestones are effectively 32x smaller.
https://github.com/JiaxuanYou/graph-generation/blob/master/train.py#L218