fix tensor dim

run_speech_recognition_seq2seq_streaming.py

@@ -334,9 +334,9 @@ class DataCollatorSpeechSeq2SeqWithPadding:
             labels = labels[:, 1:]
 
         # add start of sentence token to labels + language + task
-        labels = torch.cat((torch.full_like(labels[:, 0], self.task_id).unsqueeze(0).T, labels), dim=0)
-        labels = torch.cat((torch.full_like(labels[:, 0], self.language_id).unsqueeze(0).T, labels), dim=0)
-        labels = torch.cat((torch.full_like(labels[:, 0], self.decoder_start_token_id).unsqueeze(0).T, labels), dim=0)
+        labels = torch.cat((torch.full_like(labels[:, 0], self.task_id).unsqueeze(0).T, labels), dim=-1)
+        labels = torch.cat((torch.full_like(labels[:, 0], self.language_id).unsqueeze(0).T, labels), dim=-1)
+        labels = torch.cat((torch.full_like(labels[:, 0], self.decoder_start_token_id).unsqueeze(0).T, labels), dim=-1)
 
         batch["labels"] = labels