Support torch script. (#65)

* WIP: Support torchscript.

* Minor fixes.

* Fix style issues.

* Add documentation about how to deploy a trained model.

README.md

@@ -55,7 +55,22 @@ The WER for this model is:
 
 We provide a Colab notebook to run a pre-trained TDNN LSTM CTC model:  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1kNmDXNMwREi0rZGAOIAOJo93REBuOTcd?usp=sharing)
 
+
+## Deployment with C++
+
+Once you have trained a model in icefall, you may want to deploy it with C++,
+without Python dependencies.
+
+Please refer to the documentation
+<https://icefall.readthedocs.io/en/latest/recipes/librispeech/conformer_ctc.html#deployment-with-c>
+for how to do this.
+
+We also provide a Colab notebook, showing you how to run a torch scripted model in [k2][k2] with C++.
+Please see: [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1BIGLWzS36isskMXHKcqC9ysN6pspYXs_?usp=sharing)
+
+
 [LibriSpeech_tdnn_lstm_ctc]: egs/librispeech/ASR/tdnn_lstm_ctc
 [LibriSpeech_conformer_ctc]: egs/librispeech/ASR/conformer_ctc
 [yesno]: egs/yesno/ASR
 [librispeech]: egs/librispeech/ASR
+[k2]: https://github.com/k2-fsa/k2

docs/source/recipes/librispeech/conformer_ctc.rst

@@ -20,6 +20,7 @@ In this tutorial, you will learn:
   - (2) How to start the training, either with a single GPU or multiple GPUs
   - (3) How to do decoding after training, with n-gram LM rescoring and attention decoder rescoring
   - (4) How to use a pre-trained model, provided by us
+  - (5) How to deploy your trained model in C++, without Python dependencies
 
 Data preparation
 ----------------
@@ -693,3 +694,119 @@ We do provide a colab notebook for this recipe showing how to use a pre-trained
 
 **Congratulations!** You have finished the librispeech ASR recipe with
 conformer CTC models in ``icefall``.
+
+If you want to deploy your trained model in C++, please read the following section.
+
+Deployment with C++
+-------------------
+
+This section describes how to deploy your trained model in C++, without
+Python dependencies.
+
+We assume you have run ``./prepare.sh`` and have the following directories available:
+
+.. code-block:: bash
+
+    data
+    |-- lang_bpe
+
+Also, we assume your checkpoints are saved in ``conformer_ctc/exp``.
+
+If you know that averaging 20 checkpoints starting from ``epoch-30.pt`` yields the
+lowest WER, you can run the following commands
+
+.. code-block::
+
+  $ cd egs/librispeech/ASR
+  $ ./conformer_ctc/export.py \
+      --epoch 30 \
+      --avg 20 \
+      --jit 1 \
+      --lang-dir data/lang_bpe \
+      --exp-dir conformer_ctc/exp
+
+to get a torch scripted model saved in ``conformer_ctc/exp/cpu_jit.pt``.
+
+Now you have all needed files ready. Let us compile k2 from source:
+
+.. code-block:: bash
+
+  $ cd $HOME
+  $ git clone https://github.com/k2-fsa/k2
+  $ cd k2
+  $ git checkout v2.0-pre
+
+.. CAUTION::
+
+  You have to switch to the branch ``v2.0-pre``!
+
+.. code-block:: bash
+
+  $ mkdir build-release
+  $ cd build-release
+  $ cmake -DCMAKE_BUILD_TYPE=Release ..
+  $ make -j decode
+  # You will find an executable: `./bin/decode`
+
+Now you are ready to go!
+
+To view the usage of ``./bin/decode``, run:
+
+.. code-block::
+
+  $ ./bin/decode
+
+It will show you the following message:
+
+.. code-block::
+
+  Please provide --jit_pt
+
+    (1) CTC decoding
+      ./bin/decode \
+        --use_ctc_decoding true \
+        --jit_pt <path to exported torch script pt file> \
+        --bpe_model <path to pretrained BPE model> \
+        /path/to/foo.wav \
+        /path/to/bar.wav \
+        <more wave files if any>
+    (2) HLG decoding
+      ./bin/decode \
+        --use_ctc_decoding false \
+        --jit_pt <path to exported torch script pt file> \
+        --hlg <path to HLG.pt> \
+        --word-table <path to words.txt> \
+        /path/to/foo.wav \
+        /path/to/bar.wav \
+        <more wave files if any>
+
+     --use_gpu false to use CPU
+     --use_gpu true to use GPU
+
+``./bin/decode`` supports two types of decoding at present: CTC decoding and HLG decoding.
+
+CTC decoding
+^^^^^^^^^^^^
+
+You need to provide:
+
+  - ``--jit_pt``, this is the file generated by ``conformer_ctc/export.py``. You can find it
+    in ``conformer_ctc/exp/cpu_jit.pt``.
+  - ``--bpe_model``, this is a sentence piece model generated by ``prepare.sh``. You can find
+    it in ``data/lang_bpe/bpe.model``.
+
+
+HLG decoding
+^^^^^^^^^^^^
+
+You need to provide:
+
+  - ``--jit_pt``, this is the same file as in CTC decoding.
+  - ``--hlg``, this file is generated by ``prepare.sh``. You can find it in ``data/lang_bpe/HLG.pt``.
+  - ``--word-table``, this file is generated by ``prepare.sh``. You can find it in ``data/lang_bpe/words.txt``.
+
+We do provide a Colab notebook, showing you how to run a torch scripted model in C++.
+Please see |librispeech asr conformer ctc torch script colab notebook|
+
+.. |librispeech asr conformer ctc torch script colab notebook| image:: https://colab.research.google.com/assets/colab-badge.svg
+   :target: https://colab.research.google.com/drive/1BIGLWzS36isskMXHKcqC9ysN6pspYXs_?usp=sharing

egs/librispeech/ASR/conformer_ctc/export.py

@@ -0,0 +1,165 @@
+#!/usr/bin/env python3
+#
+# Copyright 2021 Xiaomi Corporation (Author: Fangjun Kuang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script converts several saved checkpoints
+# to a single one using model averaging.
+
+import argparse
+import logging
+from pathlib import Path
+
+import torch
+from conformer import Conformer
+
+from icefall.checkpoint import average_checkpoints, load_checkpoint
+from icefall.lexicon import Lexicon
+from icefall.utils import AttributeDict, str2bool
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=34,
+        help="It specifies the checkpoint to use for decoding."
+        "Note: Epoch counts from 0.",
+    )
+
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=20,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch'. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="conformer_ctc/exp",
+        help="""It specifies the directory where all training related
+        files, e.g., checkpoints, log, etc, are saved
+        """,
+    )
+
+    parser.add_argument(
+        "--lang-dir",
+        type=str,
+        default="data/lang_bpe",
+        help="""It contains language related input files such as "lexicon.txt"
+        """,
+    )
+
+    parser.add_argument(
+        "--jit",
+        type=str2bool,
+        default=True,
+        help="""True to save a model after applying torch.jit.script.
+        """,
+    )
+
+    return parser
+
+
+def get_params() -> AttributeDict:
+    params = AttributeDict(
+        {
+            "feature_dim": 80,
+            "subsampling_factor": 4,
+            "use_feat_batchnorm": True,
+            "attention_dim": 512,
+            "nhead": 8,
+            "num_decoder_layers": 6,
+        }
+    )
+    return params
+
+
+def main():
+    args = get_parser().parse_args()
+    args.exp_dir = Path(args.exp_dir)
+    args.lang_dir = Path(args.lang_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    logging.info(params)
+
+    lexicon = Lexicon(params.lang_dir)
+    max_token_id = max(lexicon.tokens)
+    num_classes = max_token_id + 1  # +1 for the blank
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+
+    logging.info(f"device: {device}")
+
+    model = Conformer(
+        num_features=params.feature_dim,
+        nhead=params.nhead,
+        d_model=params.attention_dim,
+        num_classes=num_classes,
+        subsampling_factor=params.subsampling_factor,
+        num_decoder_layers=params.num_decoder_layers,
+        vgg_frontend=False,
+        use_feat_batchnorm=params.use_feat_batchnorm,
+    )
+    model.to(device)
+
+    if params.avg == 1:
+        load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+    else:
+        start = params.epoch - params.avg + 1
+        filenames = []
+        for i in range(start, params.epoch + 1):
+            if start >= 0:
+                filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+        logging.info(f"averaging {filenames}")
+        model.load_state_dict(average_checkpoints(filenames))
+
+    model.to("cpu")
+    model.eval()
+
+    if params.jit:
+        logging.info("Using torch.jit.script")
+        model = torch.jit.script(model)
+        filename = params.exp_dir / "cpu_jit.pt"
+        model.save(str(filename))
+        logging.info(f"Saved to {filename}")
+    else:
+        logging.info("Not using torch.jit.script")
+        # Save it using a format so that it can be loaded
+        # by :func:`load_checkpoint`
+        filename = params.exp_dir / "pretrained.pt"
+        torch.save({"model": model.state_dict()}, str(filename))
+        logging.info(f"Saved to {filename}")
+
+
+if __name__ == "__main__":
+    formatter = (
+        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    )
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/librispeech/ASR/conformer_ctc/transformer.py

@@ -236,6 +236,7 @@ class Transformer(nn.Module):
         x = nn.functional.log_softmax(x, dim=-1)  # (N, T, C)
         return x
 
+    @torch.jit.export
     def decoder_forward(
         self,
         memory: torch.Tensor,
@@ -264,11 +265,15 @@ class Transformer(nn.Module):
         """
         ys_in = add_sos(token_ids, sos_id=sos_id)
         ys_in = [torch.tensor(y) for y in ys_in]
-        ys_in_pad = pad_sequence(ys_in, batch_first=True, padding_value=eos_id)
+        ys_in_pad = pad_sequence(
+            ys_in, batch_first=True, padding_value=float(eos_id)
+        )
 
         ys_out = add_eos(token_ids, eos_id=eos_id)
         ys_out = [torch.tensor(y) for y in ys_out]
-        ys_out_pad = pad_sequence(ys_out, batch_first=True, padding_value=-1)
+        ys_out_pad = pad_sequence(
+            ys_out, batch_first=True, padding_value=float(-1)
+        )
 
         device = memory.device
         ys_in_pad = ys_in_pad.to(device)
@@ -301,6 +306,7 @@ class Transformer(nn.Module):
 
         return decoder_loss
 
+    @torch.jit.export
     def decoder_nll(
         self,
         memory: torch.Tensor,
@@ -331,11 +337,15 @@ class Transformer(nn.Module):
 
         ys_in = add_sos(token_ids, sos_id=sos_id)
         ys_in = [torch.tensor(y) for y in ys_in]
-        ys_in_pad = pad_sequence(ys_in, batch_first=True, padding_value=eos_id)
+        ys_in_pad = pad_sequence(
+            ys_in, batch_first=True, padding_value=float(eos_id)
+        )
 
         ys_out = add_eos(token_ids, eos_id=eos_id)
         ys_out = [torch.tensor(y) for y in ys_out]
-        ys_out_pad = pad_sequence(ys_out, batch_first=True, padding_value=-1)
+        ys_out_pad = pad_sequence(
+            ys_out, batch_first=True, padding_value=float(-1)
+        )
 
         device = memory.device
         ys_in_pad = ys_in_pad.to(device, dtype=torch.int64)
@@ -649,7 +659,8 @@ class PositionalEncoding(nn.Module):
         self.d_model = d_model
         self.xscale = math.sqrt(self.d_model)
         self.dropout = nn.Dropout(p=dropout)
-        self.pe = None
+        # not doing: self.pe = None because of errors thrown by torchscript
+        self.pe = torch.zeros(0, self.d_model, dtype=torch.float32)
 
     def extend_pe(self, x: torch.Tensor) -> None:
         """Extend the time t in the positional encoding if required.
@@ -666,8 +677,7 @@ class PositionalEncoding(nn.Module):
         """
         if self.pe is not None:
             if self.pe.size(1) >= x.size(1):
-                if self.pe.dtype != x.dtype or self.pe.device != x.device:
+                self.pe = self.pe.to(dtype=x.dtype, device=x.device)
-                    self.pe = self.pe.to(dtype=x.dtype, device=x.device)
                 return
         pe = torch.zeros(x.size(1), self.d_model, dtype=torch.float32)
         position = torch.arange(0, x.size(1), dtype=torch.float32).unsqueeze(1)
@@ -972,10 +982,7 @@ def add_sos(token_ids: List[List[int]], sos_id: int) -> List[List[int]]:
       Return a new list-of-list, where each sublist starts
       with SOS ID.
     """
-    ans = []
+    return [[sos_id] + utt for utt in token_ids]
-    for utt in token_ids:
-        ans.append([sos_id] + utt)
-    return ans
 
 
 def add_eos(token_ids: List[List[int]], eos_id: int) -> List[List[int]]:
@@ -992,7 +999,4 @@ def add_eos(token_ids: List[List[int]], eos_id: int) -> List[List[int]]:
       Return a new list-of-list, where each sublist ends
       with EOS ID.
     """
-    ans = []
+    return [utt + [eos_id] for utt in token_ids]
-    for utt in token_ids:
-        ans.append(utt + [eos_id])
-    return ans

egs/librispeech/ASR/prepare.sh

@@ -41,6 +41,7 @@ dl_dir=$PWD/download
 # data/lang_bpe_yyy if the array contains xxx, yyy
 vocab_sizes=(
   5000
+  500
 )
 
 # All files generated by this script are saved in "data".