marson

I chang to DataPipeline, it still doesn't work. ` ds = wds.DataPipeline( wds.SimpleShardList(self.tar_file_list), wds.split_by_node, wds.split_by_worker, wds.tarfile_to_samples(handler=wds.warn_and_continue), wds.shuffle(1000, handler=wds.warn_and_continue), wds.decode("pilrgb", handler=wds.warn_and_continue), wds.to_tuple("jpg", "txt", handler=wds.warn_and_continue), wds.map_tuple(transform, truncate_caption, handler=wds.warn_and_continue), wds.batched(batch_size), ) `

i also have the same question, do you make it clear?

maybe [here](https://github.com/ShenhanQian/SpeechDrivesTemplates/blob/f897bfb1fa736abb7012c8e0888d0e75fa067959/core/pipelines/trainer.py#L323) should also add `@torch.no_grad()`

I use your latest repo to run the code and encounter the wrong error below, so i add `@torch.no_grad()` to `reduce_tensor_dict ` and it run successfully. `python=3.8.12 torch=1.10.0+cu113` `Traceback (most...

I just change 2D landmarks to 3D, and comment the lip_sync_error