Create tempofunk.py

tempofunk.py

@@ -0,0 +1,59 @@
+import datasets
+from io import BytesIO
+import numpy as np
+
+_TAR_FILES=[
+    "data/0000.tar",
+    "data/0001.tar",
+    "data/0002.tar",
+    "data/0003.tar",
+    "data/0004.tar",
+    "data/0005.tar",
+    "data/0006.tar",
+    "data/0007.tar",
+    "data/0008.tar",
+    "data/0009.tar",
+    ]
+
+class Food101(datasets.GeneratorBasedBuilder):
+    """Food-101 Images dataset."""
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description="TMP description",
+            homepage="google it",
+            citation="lmao",
+            license="dunno, tbh, assume the worst, k thx."
+        )
+
+    def _split_generators(self, dl_manager):
+        
+        l=[]
+        for k in _TAR_FILES:
+            archive_path = dl_manager.download(k)
+            l.append(
+                datasets.SplitGenerator(
+                name=k,
+                gen_kwargs={
+                    "npy_files": dl_manager.iter_archive(archive_path),
+                },)
+            )
+            
+        return l
+
+    def _generate_examples(self, npy_files):
+        """Generate images and labels for splits."""
+        for file_path, file_obj in npy_files:
+            # NOTE: File object is (ALREADY) opened in binary mode.
+            numpy_bytes = file_obj.read()
+            numpy_dict = np.load(BytesIO(numpy_bytes), allow_pickle=True)
+
+            reconverted_dict = {
+                "frames": numpy_dict.item().get("frames"),
+                "prompt": numpy_dict.item().get("prompt")
+            }
+
+            yield file_path, {
+                "tokenized_prompt": reconverted_dict['prompt'],
+                "video": reconverted_dict['frames']
+                }