[Fix] s2t

paddlespeech/s2t/io/dataloader.py

@@ -404,6 +404,12 @@ class DataLoaderFactory():
                 config['subsampling_factor'] = 1
                 config['num_encs'] = 1
                 config['shortest_first'] = False
+                config['minibatches'] = 0
+                config['batch_count'] = 'auto'
+                config['batch_bins'] = 0
+                config['batch_frames_in'] = 0
+                config['batch_frames_out'] = 0
+                config['batch_frames_inout'] = 0
             elif mode == 'valid':
                 config['manifest'] = config.dev_manifest
                 config['train_mode'] = False

paddlespeech/s2t/models/u2_st/u2_st.py

@@ -170,8 +170,8 @@ class U2STBaseModel(nn.Layer):
         ys_in_lens = ys_pad_lens + 1
 
         # 1. Forward decoder
-        decoder_out, _ = self.st_decoder(encoder_out, encoder_mask, ys_in_pad,
-                                         ys_in_lens)
+        decoder_out, *_ = self.st_decoder(encoder_out, encoder_mask, ys_in_pad,
+                                          ys_in_lens)
 
         # 2. Compute attention loss
         loss_att = self.criterion_att(decoder_out, ys_out_pad)
@@ -203,8 +203,8 @@ class U2STBaseModel(nn.Layer):
         ys_in_lens = ys_pad_lens + 1
 
         # 1. Forward decoder
-        decoder_out, _ = self.decoder(encoder_out, encoder_mask, ys_in_pad,
-                                      ys_in_lens)
+        decoder_out, *_ = self.decoder(encoder_out, encoder_mask, ys_in_pad,
+                                       ys_in_lens)
 
         # 2. Compute attention loss
         loss_att = self.criterion_att(decoder_out, ys_out_pad)

paddlespeech/s2t/modules/decoder.py

@@ -110,14 +110,14 @@ class TransformerDecoder(BatchScorerInterface, nn.Layer):
                 concat_after=concat_after, ) for _ in range(num_blocks)
         ])
 
-    def forward(
-            self,
-            memory: paddle.Tensor,
-            memory_mask: paddle.Tensor,
-            ys_in_pad: paddle.Tensor,
-            ys_in_lens: paddle.Tensor,
-            r_ys_in_pad: paddle.Tensor=paddle.empty([0]),
-            reverse_weight: float=0.0) -> Tuple[paddle.Tensor, paddle.Tensor]:
+    def forward(self,
+                memory: paddle.Tensor,
+                memory_mask: paddle.Tensor,
+                ys_in_pad: paddle.Tensor,
+                ys_in_lens: paddle.Tensor,
+                r_ys_in_pad: paddle.Tensor=paddle.empty([0]),
+                reverse_weight: float=0.0
+                ) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
         """Forward decoder.
         Args:
             memory: encoded memory, float32  (batch, maxlen_in, feat)