fix dp init

paddlespeech/s2t/models/u2/u2.py

@@ -605,8 +605,8 @@ class U2BaseModel(ASRInterface, nn.Layer):
             xs: paddle.Tensor,
             offset: int,
             required_cache_size: int,
-            att_cache: paddle.Tensor,  # paddle.zeros([0, 0, 0, 0])
-            cnn_cache: paddle.Tensor,  # paddle.zeros([0, 0, 0, 0])
+            att_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0]),
+            cnn_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0])
     ) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
         """ Export interface for c++ call, give input chunk xs, and return
             output from time 0 to current chunk.

paddlespeech/s2t/modules/attention.py

@@ -86,7 +86,7 @@ class MultiHeadedAttention(nn.Layer):
             self,
             value: paddle.Tensor,
             scores: paddle.Tensor,
-            mask: paddle.Tensor,  # paddle.ones([0, 0, 0], dtype=paddle.bool)
+            mask: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool)
     ) -> paddle.Tensor:
         """Compute attention context vector.
         Args:
@@ -127,15 +127,14 @@ class MultiHeadedAttention(nn.Layer):
 
         return self.linear_out(x)  # (batch, time1, d_model)
 
-    def forward(
-            self,
-            query: paddle.Tensor,
-            key: paddle.Tensor,
-            value: paddle.Tensor,
-            mask: paddle.Tensor,  # paddle.ones([0,0,0], dtype=paddle.bool)
-            pos_emb: paddle.Tensor,  # paddle.empty([0])
-            cache: paddle.Tensor  # paddle.zeros([0,0,0,0])
-    ) -> Tuple[paddle.Tensor, paddle.Tensor]:
+    def forward(self,
+                query: paddle.Tensor,
+                key: paddle.Tensor,
+                value: paddle.Tensor,
+                mask: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool),
+                pos_emb: paddle.Tensor=paddle.empty([0]),
+                cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0])
+                ) -> Tuple[paddle.Tensor, paddle.Tensor]:
         """Compute scaled dot product attention.
        Args:
             query (paddle.Tensor): Query tensor (#batch, time1, size).
@@ -244,15 +243,14 @@ class RelPositionMultiHeadedAttention(MultiHeadedAttention):
 
         return x
 
-    def forward(
-            self,
-            query: paddle.Tensor,
-            key: paddle.Tensor,
-            value: paddle.Tensor,
-            mask: paddle.Tensor,  # paddle.ones([0,0,0], dtype=paddle.bool)
-            pos_emb: paddle.Tensor,  # paddle.empty([0])
-            cache: paddle.Tensor  # paddle.zeros([0,0,0,0])
-    ) -> Tuple[paddle.Tensor, paddle.Tensor]:
+    def forward(self,
+                query: paddle.Tensor,
+                key: paddle.Tensor,
+                value: paddle.Tensor,
+                mask: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool),
+                pos_emb: paddle.Tensor=paddle.empty([0]),
+                cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0])
+                ) -> Tuple[paddle.Tensor, paddle.Tensor]:
         """Compute 'Scaled Dot Product Attention' with rel. positional encoding.
         Args:
             query (paddle.Tensor): Query tensor (#batch, time1, size).

paddlespeech/s2t/modules/conformer_convolution.py

@@ -108,8 +108,8 @@ class ConvolutionModule(nn.Layer):
     def forward(
             self,
             x: paddle.Tensor,
-            mask_pad: paddle.Tensor,  # paddle.ones([0,0,0], dtype=paddle.bool)
-            cache: paddle.Tensor  # paddle.zeros([0,0,0,0])
+            mask_pad: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool),
+            cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0])
     ) -> Tuple[paddle.Tensor, paddle.Tensor]:
         """Compute convolution module.
         Args:

paddlespeech/s2t/modules/decoder_layer.py

@@ -121,16 +121,11 @@ class DecoderLayer(nn.Layer):
 
         if self.concat_after:
             tgt_concat = paddle.cat(
-                (tgt_q, self.self_attn(tgt_q, tgt, tgt, tgt_q_mask,
-                                       paddle.empty([0]),
-                                       paddle.zeros([0, 0, 0, 0]))[0]),
-                dim=-1)
+                (tgt_q, self.self_attn(tgt_q, tgt, tgt, tgt_q_mask)[0]), dim=-1)
             x = residual + self.concat_linear1(tgt_concat)
         else:
             x = residual + self.dropout(
-                self.self_attn(tgt_q, tgt, tgt, tgt_q_mask,
-                               paddle.empty([0]), paddle.zeros([0, 0, 0, 0]))[
-                                   0])
+                self.self_attn(tgt_q, tgt, tgt, tgt_q_mask)[0])
         if not self.normalize_before:
             x = self.norm1(x)
 
@@ -139,15 +134,11 @@ class DecoderLayer(nn.Layer):
             x = self.norm2(x)
         if self.concat_after:
             x_concat = paddle.cat(
-                (x, self.src_attn(x, memory, memory, memory_mask,
-                                  paddle.empty([0]),
-                                  paddle.zeros([0, 0, 0, 0]))[0]),
-                dim=-1)
+                (x, self.src_attn(x, memory, memory, memory_mask)[0]), dim=-1)
             x = residual + self.concat_linear2(x_concat)
         else:
             x = residual + self.dropout(
-                self.src_attn(x, memory, memory, memory_mask,
-                              paddle.empty([0]), paddle.zeros([0, 0, 0, 0]))[0])
+                self.src_attn(x, memory, memory, memory_mask)[0])
         if not self.normalize_before:
             x = self.norm2(x)
 

paddlespeech/s2t/modules/encoder.py

@@ -175,9 +175,7 @@ class BaseEncoder(nn.Layer):
             decoding_chunk_size, self.static_chunk_size,
             num_decoding_left_chunks)
         for layer in self.encoders:
-            xs, chunk_masks, _, _ = layer(xs, chunk_masks, pos_emb, mask_pad,
-                                          paddle.zeros([0, 0, 0, 0]),
-                                          paddle.zeros([0, 0, 0, 0]))
+            xs, chunk_masks, _, _ = layer(xs, chunk_masks, pos_emb, mask_pad)
         if self.normalize_before:
             xs = self.after_norm(xs)
         # Here we assume the mask is not changed in encoder layers, so just
@@ -190,9 +188,9 @@ class BaseEncoder(nn.Layer):
             xs: paddle.Tensor,
             offset: int,
             required_cache_size: int,
-            att_cache: paddle.Tensor,  # paddle.zeros([0,0,0,0])
-            cnn_cache: paddle.Tensor,  # paddle.zeros([0,0,0,0]),
-            att_mask: paddle.Tensor,  # paddle.ones([0,0,0], dtype=paddle.bool)
+            att_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0]),
+            cnn_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0]),
+            att_mask: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool)
     ) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
         """ Forward just one chunk
         Args:
@@ -255,7 +253,6 @@ class BaseEncoder(nn.Layer):
                 xs,
                 att_mask,
                 pos_emb,
-                mask_pad=paddle.ones([0, 0, 0], dtype=paddle.bool),
                 att_cache=att_cache[i:i + 1] if elayers > 0 else att_cache,
                 cnn_cache=cnn_cache[i:i + 1]
                 if paddle.shape(cnn_cache)[0] > 0 else cnn_cache, )
@@ -328,8 +325,7 @@ class BaseEncoder(nn.Layer):
             chunk_xs = xs[:, cur:end, :]
 
             (y, att_cache, cnn_cache) = self.forward_chunk(
-                chunk_xs, offset, required_cache_size, att_cache, cnn_cache,
-                paddle.ones([0, 0, 0], dtype=paddle.bool))
+                chunk_xs, offset, required_cache_size, att_cache, cnn_cache)
 
             outputs.append(y)
             offset += y.shape[1]

paddlespeech/s2t/modules/encoder_layer.py

@@ -76,10 +76,9 @@ class TransformerEncoderLayer(nn.Layer):
             x: paddle.Tensor,
             mask: paddle.Tensor,
             pos_emb: paddle.Tensor,
-            mask_pad: paddle.
-            Tensor,  # paddle.ones([0, 0, 0], dtype=paddle.bool)
-            att_cache: paddle.Tensor,  # paddle.zeros([0, 0, 0, 0])
-            cnn_cache: paddle.Tensor,  # paddle.zeros([0, 0, 0, 0])
+            mask_pad: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool),
+            att_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0]),
+            cnn_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0])
     ) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor, paddle.Tensor]:
         """Compute encoded features.
         Args:
@@ -106,8 +105,7 @@ class TransformerEncoderLayer(nn.Layer):
         if self.normalize_before:
             x = self.norm1(x)
 
-        x_att, new_att_cache = self.self_attn(
-            x, x, x, mask, paddle.empty([0]), cache=att_cache)
+        x_att, new_att_cache = self.self_attn(x, x, x, mask, cache=att_cache)
 
         if self.concat_after:
             x_concat = paddle.concat((x, x_att), axis=-1)
@@ -195,9 +193,9 @@ class ConformerEncoderLayer(nn.Layer):
             x: paddle.Tensor,
             mask: paddle.Tensor,
             pos_emb: paddle.Tensor,
-            mask_pad: paddle.Tensor,  #paddle.ones([0, 0, 0],dtype=paddle.bool)
-            att_cache: paddle.Tensor,  # paddle.zeros([0, 0, 0, 0])
-            cnn_cache: paddle.Tensor,  # paddle.zeros([0, 0, 0, 0])
+            mask_pad: paddle.Tensor=paddle.ones([0, 0, 0], dtype=paddle.bool),
+            att_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0]),
+            cnn_cache: paddle.Tensor=paddle.zeros([0, 0, 0, 0])
     ) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor, paddle.Tensor]:
         """Compute encoded features.
         Args:

paddlespeech/server/engine/asr/online/python/asr_engine.py

@@ -480,8 +480,7 @@ class PaddleASRConnectionHanddler:
                  self.offset,
                  required_cache_size,
                  att_cache=self.att_cache,
-                 cnn_cache=self.cnn_cache,
-                 att_mask=paddle.ones([0, 0, 0], dtype=paddle.bool))
+                 cnn_cache=self.cnn_cache)
             outputs.append(y)
 
             # update the global offset, in decoding frame unit