Export stateless pruned transducer to jit script module

egs/librispeech/ASR/pruned_transducer_stateless/export.py

@@ -116,8 +116,6 @@ def main():
     args = get_parser().parse_args()
     args.exp_dir = Path(args.exp_dir)
 
-    assert args.jit is False, "Support torchscript will be added later"
-
     params = get_params()
     params.update(vars(args))
 

egs/librispeech/ASR/pruned_transducer_stateless/model.py

@@ -15,6 +15,8 @@
 # limitations under the License.
 
 
+from typing import Tuple
+
 import k2
 import torch
 import torch.nn as nn
@@ -58,6 +60,12 @@ class Transducer(nn.Module):
         self.decoder = decoder
         self.joiner = joiner
 
+        self.subsampling_factor = self.encoder.subsampling_factor
+        self.context_size = self.decoder.context_size
+        self.blank_id = self.decoder.blank_id
+        self.vocab_size = self.decoder.vocab_size
+
+    @torch.jit.ignore
     def forward(
         self,
         x: torch.Tensor,
@@ -167,3 +175,49 @@ class Transducer(nn.Module):
         )
 
         return (simple_loss, pruned_loss)
+
+    @torch.jit.export
+    def encoder_forward(
+        self, x: torch.Tensor, x_lens: torch.Tensor
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Args:
+          x:
+            The input tensor. Its shape is (batch_size, seq_len, feature_dim).
+          x_lens:
+            A tensor of shape (batch_size,) containing the number of frames in
+            `x` before padding.
+        Returns:
+          Return a tuple containing 2 tensors:
+            - logits, its shape is (batch_size, output_seq_len, output_dim)
+            - logit_lens, a tensor of shape (batch_size,) containing the number
+              of frames in `logits` before padding.
+        """
+        return self.encoder(x, x_lens)
+
+    @torch.jit.export
+    def decoder_forward(self, y: torch.Tensor) -> torch.Tensor:
+        """
+        Args:
+          y:
+            A 2-D tensor of shape (N, U) with blank prepended.
+        Returns:
+          Return a tensor of shape (N, U, embedding_dim).
+        """
+        return self.decoder(y, False)
+
+    @torch.jit.export
+    def joiner_forward(
+        self, encoder_out: torch.Tensor, decoder_out: torch.Tensor
+    ) -> torch.Tensor:
+        """
+        Args:
+          encoder_out:
+            Output from the encoder. Its shape is (N, 1, C).
+          decoder_out:
+            Output from the decoder. Its shape is (N, 1, C).
+        Returns:
+          Return a tensor of shape (N, C).
+        """
+        logits = self.joiner(encoder_out.unsqueeze(2), decoder_out.unsqueeze(1))
+        return logits.squeeze(1).squeeze(1)