[fbsync] Remove split parameter from datasets.load() (#5207)

Summary:
* Maybe, Maybe not

* add test

* cleanup

* Revert "cleanup"

This reverts commit aa4285f.

* Revert "add test"

This reverts commit 1f8a9cc.

* cleanup

Reviewed By: jdsgomes, prabhat00155

Differential Revision: D33739377

fbshipit-source-id: 2da0e4d3c56fd50dd51139cd062126d1e86f600a

Co-authored-by: Philip Meier <[email protected]>

Author: 2 people

torchvision/prototype/datasets/_api.py

@@ -62,15 +62,14 @@ def load(
     *,
     decoder: Optional[Callable[[io.IOBase], torch.Tensor]] = DEFAULT_DECODER,  # type: ignore[assignment]
     skip_integrity_check: bool = False,
-    split: str = "train",
     **options: Any,
 ) -> IterDataPipe[Dict[str, Any]]:
     dataset = find(name)
 
     if decoder is DEFAULT_DECODER:
         decoder = DEFAULT_DECODER_MAP.get(dataset.info.type)
 
-    config = dataset.info.make_config(split=split, **options)
+    config = dataset.info.make_config(**options)
     root = os.path.join(home(), dataset.name)
 
     return dataset.load(root, config=config, decoder=decoder, skip_integrity_check=skip_integrity_check)