dataset builder?

commavq.py

@@ -0,0 +1,36 @@
+import datasets
+from zipfile import ZipFile
+import numpy as np
+
+SHARD_SIZE = 2500
+NUM_SHARDS = 40
+_URLS = [
+  f'https://huggingface.co/datasets/commaai/commavq/blob/main/data_{i*SHARD_SIZE}_to_{(i+1)*SHARD_SIZE}.zip' for i in range(NUM_SHARDS)
+  ]
+
+_DESCRIPTION = """\
+TODO
+"""
+
+class CommaVQ(datasets.GeneratorBasedBuilder):
+  def _info(self):
+
+    return datasets.DatasetInfo(
+      # This is the description that will appear on the datasets page.
+      description=_DESCRIPTION,
+    )
+
+  def _split_generators(self, dl_manager):
+    """Returns SplitGenerators."""
+    downloaded_files = dl_manager.download_and_extract(_URLS)
+
+    return [
+      datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": x}) for x in downloaded_files
+    ]
+
+  def _generate_examples(self, filepaths):
+    """Yields examples."""
+    for filepath in filepaths:
+      input_zip = ZipFile(filepath)
+      for name in input_zip.namelist():
+        yield name, {'tokens': input_zip.read(name)}