train on all data

run_speech_recognition_seq2seq_streaming.py

@@ -453,7 +453,7 @@ def load_maybe_streaming_dataset(
                 dataset_splits.append(dataset)
 
         # interleave multiple splits to form one dataset
-        interleaved_dataset = interleave_datasets(dataset_splits)
+        interleaved_dataset = interleave_datasets(dataset_splits, stopping_strategy="all_exhausted")
         return interleaved_dataset
     else:
         # load a single split *with* streaming mode