From 45d60ef262fe65fa1a63cbdd7b89658b359f7724 Mon Sep 17 00:00:00 2001 From: l2009312042 Date: Thu, 21 Sep 2023 19:41:10 +0800 Subject: [PATCH 1/3] Update conformer.py (#1200) * Update conformer.py * Update zipformer.py fix bug in get_dynamic_dropout_rate --- .../ASR/pruned_transducer_stateless7_streaming/zipformer.py | 2 +- egs/librispeech/ASR/streaming_conformer_ctc/conformer.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/zipformer.py b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/zipformer.py index a5c422959..c7e45564f 100644 --- a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/zipformer.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/zipformer.py @@ -865,7 +865,7 @@ class ZipformerEncoderLayer(nn.Module): return final_dropout_rate else: return initial_dropout_rate - ( - initial_dropout_rate * final_dropout_rate + initial_dropout_rate - final_dropout_rate ) * (self.batch_count / warmup_period) def forward( diff --git a/egs/librispeech/ASR/streaming_conformer_ctc/conformer.py b/egs/librispeech/ASR/streaming_conformer_ctc/conformer.py index 5fe92172e..be6fabf35 100644 --- a/egs/librispeech/ASR/streaming_conformer_ctc/conformer.py +++ b/egs/librispeech/ASR/streaming_conformer_ctc/conformer.py @@ -230,7 +230,7 @@ class Conformer(Transformer): x, pos_emb, mask=mask, src_key_padding_mask=src_key_padding_mask ) # (T, B, F) else: - x = self.encoder(x, pos_emb, src_key_padding_mask=mask) # (T, B, F) + x = self.encoder(x, pos_emb, src_key_padding_mask=src_key_padding_mask) # (T, B, F) if self.normalize_before: x = self.after_norm(x) From f5dc957d44350ea0ec9adb81578c32af5e6bb809 Mon Sep 17 00:00:00 2001 From: Fangjun Kuang Date: Thu, 21 Sep 2023 21:16:14 +0800 Subject: [PATCH 2/3] Fix CI tests (#1266) --- .../ASR/pruned_transducer_stateless7/onnx_pretrained.py | 3 +++ .../ASR/pruned_transducer_stateless7/onnx_pretrained.py | 3 +++ .../onnx_pretrained.py | 3 +++ .../ASR/lstm_transducer_stateless2/onnx_pretrained.py | 3 +++ .../lstm_transducer_stateless2/streaming-onnx-decode.py | 5 +++++ .../ASR/pruned_transducer_stateless3/onnx_pretrained.py | 3 +++ .../ASR/pruned_transducer_stateless3/test_onnx.py | 5 +++++ .../onnx_pretrained-streaming.py | 3 +++ .../ASR/pruned_transducer_stateless7/test_onnx.py | 5 +++++ .../onnx_pretrained.py | 8 ++++++++ .../onnx_pretrained.py | 3 +++ .../ASR/zipformer/onnx_pretrained-streaming.py | 3 +++ egs/librispeech/ASR/zipformer/onnx_pretrained.py | 3 +++ .../ASR/pruned_transducer_stateless2/onnx_check.py | 5 +++++ .../onnx_pretrained-streaming.py | 3 +++ .../ASR/pruned_transducer_stateless5/onnx_pretrained.py | 3 +++ egs/yesno/ASR/tdnn/onnx_pretrained.py | 1 + 17 files changed, 62 insertions(+) diff --git a/egs/aishell/ASR/pruned_transducer_stateless7/onnx_pretrained.py b/egs/aishell/ASR/pruned_transducer_stateless7/onnx_pretrained.py index 5adb6c16a..a92182e8d 100755 --- a/egs/aishell/ASR/pruned_transducer_stateless7/onnx_pretrained.py +++ b/egs/aishell/ASR/pruned_transducer_stateless7/onnx_pretrained.py @@ -151,12 +151,14 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_decoder(self, decoder_model_filename: str): self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -170,6 +172,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/commonvoice/ASR/pruned_transducer_stateless7/onnx_pretrained.py b/egs/commonvoice/ASR/pruned_transducer_stateless7/onnx_pretrained.py index eee19191e..cf6ddfa36 100755 --- a/egs/commonvoice/ASR/pruned_transducer_stateless7/onnx_pretrained.py +++ b/egs/commonvoice/ASR/pruned_transducer_stateless7/onnx_pretrained.py @@ -152,12 +152,14 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_decoder(self, decoder_model_filename: str): self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -171,6 +173,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/conv_emformer_transducer_stateless2/onnx_pretrained.py b/egs/librispeech/ASR/conv_emformer_transducer_stateless2/onnx_pretrained.py index 5d7e2dfcd..a6c69d54f 100755 --- a/egs/librispeech/ASR/conv_emformer_transducer_stateless2/onnx_pretrained.py +++ b/egs/librispeech/ASR/conv_emformer_transducer_stateless2/onnx_pretrained.py @@ -136,6 +136,7 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) self.init_encoder_states() @@ -184,6 +185,7 @@ class OnnxModel: self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -197,6 +199,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/lstm_transducer_stateless2/onnx_pretrained.py b/egs/librispeech/ASR/lstm_transducer_stateless2/onnx_pretrained.py index fb9e121e5..06159e56a 100755 --- a/egs/librispeech/ASR/lstm_transducer_stateless2/onnx_pretrained.py +++ b/egs/librispeech/ASR/lstm_transducer_stateless2/onnx_pretrained.py @@ -129,6 +129,7 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) self.init_encoder_states() @@ -166,6 +167,7 @@ class OnnxModel: self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -179,6 +181,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/lstm_transducer_stateless2/streaming-onnx-decode.py b/egs/librispeech/ASR/lstm_transducer_stateless2/streaming-onnx-decode.py index 34d2e5630..487fc2114 100755 --- a/egs/librispeech/ASR/lstm_transducer_stateless2/streaming-onnx-decode.py +++ b/egs/librispeech/ASR/lstm_transducer_stateless2/streaming-onnx-decode.py @@ -172,30 +172,35 @@ class Model: self.encoder = ort.InferenceSession( args.encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_decoder(self, args): self.decoder = ort.InferenceSession( args.decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_joiner(self, args): self.joiner = ort.InferenceSession( args.joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_joiner_encoder_proj(self, args): self.joiner_encoder_proj = ort.InferenceSession( args.joiner_encoder_proj_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_joiner_decoder_proj(self, args): self.joiner_decoder_proj = ort.InferenceSession( args.joiner_decoder_proj_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def run_encoder(self, x, h0, c0) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: diff --git a/egs/librispeech/ASR/pruned_transducer_stateless3/onnx_pretrained.py b/egs/librispeech/ASR/pruned_transducer_stateless3/onnx_pretrained.py index e10915086..de3e03da6 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless3/onnx_pretrained.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless3/onnx_pretrained.py @@ -150,12 +150,14 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_decoder(self, decoder_model_filename: str): self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -169,6 +171,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/pruned_transducer_stateless3/test_onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless3/test_onnx.py index 810da8da6..b98248128 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless3/test_onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless3/test_onnx.py @@ -78,6 +78,7 @@ def test_conv2d_subsampling(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -133,6 +134,7 @@ def test_rel_pos(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -220,6 +222,7 @@ def test_conformer_encoder_layer(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -304,6 +307,7 @@ def test_conformer_encoder(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -359,6 +363,7 @@ def test_conformer(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() diff --git a/egs/librispeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py b/egs/librispeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py index 29be4c655..6e290e799 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py @@ -138,6 +138,7 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) self.init_encoder_states() @@ -185,6 +186,7 @@ class OnnxModel: self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -198,6 +200,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7/test_onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless7/test_onnx.py index 1e9b67226..f3f7b1ea9 100644 --- a/egs/librispeech/ASR/pruned_transducer_stateless7/test_onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7/test_onnx.py @@ -74,6 +74,7 @@ def test_conv2d_subsampling(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -128,6 +129,7 @@ def test_rel_pos(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -204,6 +206,7 @@ def test_zipformer_encoder_layer(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -284,6 +287,7 @@ def test_zipformer_encoder(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() @@ -338,6 +342,7 @@ def test_zipformer(): session = ort.InferenceSession( filename, sess_options=options, + providers=["CPUExecutionProvider"], ) input_nodes = session.get_inputs() diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7_ctc_bs/onnx_pretrained.py b/egs/librispeech/ASR/pruned_transducer_stateless7_ctc_bs/onnx_pretrained.py index 8ff02fbcb..494a34d97 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless7_ctc_bs/onnx_pretrained.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7_ctc_bs/onnx_pretrained.py @@ -326,41 +326,49 @@ def main(): encoder = ort.InferenceSession( args.encoder_model_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) decoder = ort.InferenceSession( args.decoder_model_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) joiner = ort.InferenceSession( args.joiner_model_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) joiner_encoder_proj = ort.InferenceSession( args.joiner_encoder_proj_model_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) joiner_decoder_proj = ort.InferenceSession( args.joiner_decoder_proj_model_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) lconv = ort.InferenceSession( args.lconv_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) frame_reducer = ort.InferenceSession( args.frame_reducer_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) ctc_output = ort.InferenceSession( args.ctc_output_filename, sess_options=session_opts, + providers=["CPUExecutionProvider"], ) sp = spm.SentencePieceProcessor() diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/onnx_pretrained.py b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/onnx_pretrained.py index 8192e01fd..04861ea37 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/onnx_pretrained.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/onnx_pretrained.py @@ -130,6 +130,7 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) self.init_encoder_states() @@ -229,6 +230,7 @@ class OnnxModel: self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -242,6 +244,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/zipformer/onnx_pretrained-streaming.py b/egs/librispeech/ASR/zipformer/onnx_pretrained-streaming.py index 500b2cd09..e62491444 100755 --- a/egs/librispeech/ASR/zipformer/onnx_pretrained-streaming.py +++ b/egs/librispeech/ASR/zipformer/onnx_pretrained-streaming.py @@ -146,6 +146,7 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) self.init_encoder_states() @@ -236,6 +237,7 @@ class OnnxModel: self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -249,6 +251,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/librispeech/ASR/zipformer/onnx_pretrained.py b/egs/librispeech/ASR/zipformer/onnx_pretrained.py index 032b07721..334376093 100755 --- a/egs/librispeech/ASR/zipformer/onnx_pretrained.py +++ b/egs/librispeech/ASR/zipformer/onnx_pretrained.py @@ -151,12 +151,14 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_decoder(self, decoder_model_filename: str): self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -170,6 +172,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless2/onnx_check.py b/egs/wenetspeech/ASR/pruned_transducer_stateless2/onnx_check.py index a46ff5a07..2d46eede1 100755 --- a/egs/wenetspeech/ASR/pruned_transducer_stateless2/onnx_check.py +++ b/egs/wenetspeech/ASR/pruned_transducer_stateless2/onnx_check.py @@ -258,6 +258,7 @@ def main(): encoder_session = ort.InferenceSession( args.onnx_encoder_filename, sess_options=options, + providers=["CPUExecutionProvider"], ) test_encoder(model, encoder_session) @@ -265,6 +266,7 @@ def main(): decoder_session = ort.InferenceSession( args.onnx_decoder_filename, sess_options=options, + providers=["CPUExecutionProvider"], ) test_decoder(model, decoder_session) @@ -272,14 +274,17 @@ def main(): joiner_session = ort.InferenceSession( args.onnx_joiner_filename, sess_options=options, + providers=["CPUExecutionProvider"], ) joiner_encoder_proj_session = ort.InferenceSession( args.onnx_joiner_encoder_proj_filename, sess_options=options, + providers=["CPUExecutionProvider"], ) joiner_decoder_proj_session = ort.InferenceSession( args.onnx_joiner_decoder_proj_filename, sess_options=options, + providers=["CPUExecutionProvider"], ) test_joiner( model, diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py b/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py index facfc2258..c31db6859 100755 --- a/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py +++ b/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained-streaming.py @@ -139,6 +139,7 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) self.init_encoder_states() @@ -186,6 +187,7 @@ class OnnxModel: self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -199,6 +201,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained.py b/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained.py index e7c8b4556..c784853ee 100755 --- a/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained.py +++ b/egs/wenetspeech/ASR/pruned_transducer_stateless5/onnx_pretrained.py @@ -158,12 +158,14 @@ class OnnxModel: self.encoder = ort.InferenceSession( encoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) def init_decoder(self, decoder_model_filename: str): self.decoder = ort.InferenceSession( decoder_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) decoder_meta = self.decoder.get_modelmeta().custom_metadata_map @@ -177,6 +179,7 @@ class OnnxModel: self.joiner = ort.InferenceSession( joiner_model_filename, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) joiner_meta = self.joiner.get_modelmeta().custom_metadata_map diff --git a/egs/yesno/ASR/tdnn/onnx_pretrained.py b/egs/yesno/ASR/tdnn/onnx_pretrained.py index b23a2a381..72a1d69c8 100755 --- a/egs/yesno/ASR/tdnn/onnx_pretrained.py +++ b/egs/yesno/ASR/tdnn/onnx_pretrained.py @@ -54,6 +54,7 @@ class OnnxModel: self.model = ort.InferenceSession( nn_model, sess_options=self.session_opts, + providers=["CPUExecutionProvider"], ) meta = self.model.get_modelmeta().custom_metadata_map From 34e40a86b33102576b3442329421178a487e3ea3 Mon Sep 17 00:00:00 2001 From: Fangjun Kuang Date: Fri, 22 Sep 2023 09:57:15 +0800 Subject: [PATCH 3/3] Fix exporting decoder model to onnx (#1264) * Use torch.jit.script() to export the decoder model See also https://github.com/k2-fsa/sherpa-onnx/issues/327 --- egs/aishell/ASR/pruned_transducer_stateless7/export-onnx.py | 1 + egs/commonvoice/ASR/pruned_transducer_stateless7/export-onnx.py | 1 + .../ASR/conv_emformer_transducer_stateless2/export-onnx.py | 1 + egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx-zh.py | 1 + egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx.py | 1 + egs/librispeech/ASR/pruned_transducer_stateless/export-onnx.py | 1 + egs/librispeech/ASR/pruned_transducer_stateless3/export-onnx.py | 1 + .../ASR/pruned_transducer_stateless5/export-onnx-streaming.py | 1 + egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx.py | 1 + egs/librispeech/ASR/pruned_transducer_stateless7/export-onnx.py | 1 + .../ASR/pruned_transducer_stateless7_streaming/export-onnx-zh.py | 1 + .../ASR/pruned_transducer_stateless7_streaming/export-onnx.py | 1 + egs/librispeech/ASR/zipformer/export-onnx-streaming.py | 1 + egs/librispeech/ASR/zipformer/export-onnx.py | 1 + egs/wenetspeech/ASR/pruned_transducer_stateless2/export-onnx.py | 1 + .../ASR/pruned_transducer_stateless5/export-onnx-streaming.py | 1 + egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx.py | 1 + 17 files changed, 17 insertions(+) diff --git a/egs/aishell/ASR/pruned_transducer_stateless7/export-onnx.py b/egs/aishell/ASR/pruned_transducer_stateless7/export-onnx.py index e8211500a..2a9fc57d5 100755 --- a/egs/aishell/ASR/pruned_transducer_stateless7/export-onnx.py +++ b/egs/aishell/ASR/pruned_transducer_stateless7/export-onnx.py @@ -322,6 +322,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/commonvoice/ASR/pruned_transducer_stateless7/export-onnx.py b/egs/commonvoice/ASR/pruned_transducer_stateless7/export-onnx.py index 0c98885ac..2b9f2293a 100755 --- a/egs/commonvoice/ASR/pruned_transducer_stateless7/export-onnx.py +++ b/egs/commonvoice/ASR/pruned_transducer_stateless7/export-onnx.py @@ -330,6 +330,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/conv_emformer_transducer_stateless2/export-onnx.py b/egs/librispeech/ASR/conv_emformer_transducer_stateless2/export-onnx.py index cfd365207..ab046557f 100755 --- a/egs/librispeech/ASR/conv_emformer_transducer_stateless2/export-onnx.py +++ b/egs/librispeech/ASR/conv_emformer_transducer_stateless2/export-onnx.py @@ -401,6 +401,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx-zh.py b/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx-zh.py index 89ced388c..2a52e2eec 100755 --- a/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx-zh.py +++ b/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx-zh.py @@ -359,6 +359,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx.py b/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx.py index 6b6cb893f..c543628ff 100755 --- a/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx.py +++ b/egs/librispeech/ASR/lstm_transducer_stateless2/export-onnx.py @@ -356,6 +356,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless/export-onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless/export-onnx.py index 282238c13..0a2132e56 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless/export-onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless/export-onnx.py @@ -307,6 +307,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless3/export-onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless3/export-onnx.py index 26dea7e11..2685ea95a 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless3/export-onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless3/export-onnx.py @@ -312,6 +312,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py b/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py index 549fb13c9..b90d81dcf 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py @@ -404,6 +404,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx.py index fff0fcdd5..02aa24f2c 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless5/export-onnx.py @@ -335,6 +335,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7/export-onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless7/export-onnx.py index 11c885f4d..b75548f8b 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless7/export-onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7/export-onnx.py @@ -329,6 +329,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx-zh.py b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx-zh.py index 8653126de..2de56837e 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx-zh.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx-zh.py @@ -413,6 +413,7 @@ def export_decoder_model_onnx( context_size = decoder_model.decoder.context_size vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx.py b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx.py index 6f84d79b4..d71080760 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/export-onnx.py @@ -401,6 +401,7 @@ def export_decoder_model_onnx( context_size = decoder_model.decoder.context_size vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/zipformer/export-onnx-streaming.py b/egs/librispeech/ASR/zipformer/export-onnx-streaming.py index a951aeef3..e2c7d7d95 100755 --- a/egs/librispeech/ASR/zipformer/export-onnx-streaming.py +++ b/egs/librispeech/ASR/zipformer/export-onnx-streaming.py @@ -506,6 +506,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/librispeech/ASR/zipformer/export-onnx.py b/egs/librispeech/ASR/zipformer/export-onnx.py index e0d664009..3682f0b62 100755 --- a/egs/librispeech/ASR/zipformer/export-onnx.py +++ b/egs/librispeech/ASR/zipformer/export-onnx.py @@ -353,6 +353,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless2/export-onnx.py b/egs/wenetspeech/ASR/pruned_transducer_stateless2/export-onnx.py index 760fad974..140b1d37f 100755 --- a/egs/wenetspeech/ASR/pruned_transducer_stateless2/export-onnx.py +++ b/egs/wenetspeech/ASR/pruned_transducer_stateless2/export-onnx.py @@ -315,6 +315,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py b/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py index 9a926d7e5..921766ad4 100755 --- a/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py +++ b/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx-streaming.py @@ -404,6 +404,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y, diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx.py b/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx.py index 68c7cc352..037c7adf1 100755 --- a/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx.py +++ b/egs/wenetspeech/ASR/pruned_transducer_stateless5/export-onnx.py @@ -335,6 +335,7 @@ def export_decoder_model_onnx( vocab_size = decoder_model.decoder.vocab_size y = torch.zeros(10, context_size, dtype=torch.int64) + decoder_model = torch.jit.script(decoder_model) torch.onnx.export( decoder_model, y,