Remove tf glob and pass list of files

P3.py

@@ -64,7 +64,7 @@ def load_cached_task(features_dict, tfrecord):
         feat: _feature_config(**desc) for feat, desc in features_dict.items()
     }
 
-    ds = tf.data.TFRecordDataset(tf.io.gfile.glob([tfrecord])) # TODO -> handle multiple shards
+    ds = tf.data.TFRecordDataset(tfrecord)
     ds = ds.map(
         lambda pb: tf.io.parse_single_example(pb, feature_description),
         num_parallel_calls=tf.data.experimental.AUTOTUNE
@@ -158,7 +158,7 @@ class P3(datasets.GeneratorBasedBuilder):
                 datasets.SplitGenerator(
                     name=datasets.Split.TRAIN,
                     gen_kwargs={
-                        "tfrecord": data_dir[split_name][0],
+                        "tfrecord": data_dir[split_name],
                     }
                 )
             )
@@ -168,7 +168,7 @@ class P3(datasets.GeneratorBasedBuilder):
                 datasets.SplitGenerator(
                     name=datasets.Split.VALIDATION,
                     gen_kwargs={
-                        "tfrecord": data_dir[split_name][0],
+                        "tfrecord": data_dir[split_name],
                     }
                 )
             )
@@ -178,7 +178,7 @@ class P3(datasets.GeneratorBasedBuilder):
                 datasets.SplitGenerator(
                     name=datasets.Split.TEST,
                     gen_kwargs={
-                        "tfrecord": data_dir[split_name][0],
+                        "tfrecord": data_dir[split_name],
                     }
                 )
             )
@@ -189,7 +189,7 @@ class P3(datasets.GeneratorBasedBuilder):
                 datasets.SplitGenerator(
                     name=datasets.Split(special_split_name),
                     gen_kwargs={
-                        "tfrecord": data_dir[special_split_name][0],
+                        "tfrecord": data_dir[special_split_name],
                     }
                 )
             )