Lifeng Chen

Results 2 issues of Lifeng Chen

DDP with Accelerate: you need to add an explicit nodesplitter to your input pipeline for multi-node training Webdataset

I‘m using my Accelerator to prepare the dataloader, which is initialized and prepared as followed: ```python tar_files = glob.glob(wds_path) dataset = wds.WebDataset(tar_files, resampled=True).decode("pil").shuffle(1000).to_tuple("jpg", "txt").with_epoch(10000) train_dataloader= DataLoader(dataset, batch_size=train_batch_size, num_workers=num_workers, collate_fn=collate_fn) ip_adapter,...

Optimization learning rate set

Dear authors, I notice that you set the scale_factor as 20, and scale_range to be (1.0,0.5); How were them determined? Could anybody give me a hand?