Add wav2letter model to examples (#71)

Summary:
Pull Request resolved: #71

Bring in `Wav2Letter` model to `executorch/examples`.

  - General info about `Wav2Letter` model: https://ai.meta.com/tools/wav2letter/
  - Info about `Wav2Letter` model being used in this example: https://pytorch.org/audio/stable/_modules/torchaudio/models/wav2letter.html

Reviewed By: JacobSzwejbka

Differential Revision: D48403704

fbshipit-source-id: 3e5e56861dc8a4a3b93744a35c3b6dc8f330c1ce

Author: guangy10

examples/export/test/test_export.py

@@ -81,3 +81,11 @@ def test_vit_export_to_executorch(self):
         self._assert_eager_lowered_same_result(
             eager_model, example_inputs, self.validate_tensor_allclose
         )
+
+    def test_w2l_export_to_executorch(self):
+        eager_model, example_inputs = MODEL_NAME_TO_MODEL["w2l"]()
+        eager_model = eager_model.eval()
+
+        self._assert_eager_lowered_same_result(
+            eager_model, example_inputs, self.validate_tensor_allclose
+        )

examples/models/TARGETS

@@ -11,6 +11,7 @@ python_library(
         "//executorch/examples/models/mobilenet_v2:mv2_export",
         "//executorch/examples/models/mobilenet_v3:mv3_export",
         "//executorch/examples/models/torchvision_vit:vit_export",
+        "//executorch/examples/models/wav2letter:w2l_export",
         "//executorch/exir/backend:compile_spec_schema",
     ],
 )

examples/models/models.py

@@ -95,6 +95,13 @@ def gen_torchvision_vit_model_and_inputs() -> Tuple[torch.nn.Module, Any]:
     return TorchVisionViTModel.get_model(), TorchVisionViTModel.get_example_inputs()
 
 
+def gen_wav2letter_model_and_inputs() -> Tuple[torch.nn.Module, Any]:
+    from ..models.wav2letter import Wav2LetterModel
+
+    model = Wav2LetterModel()
+    return model.get_model(), model.get_example_inputs()
+
+
 MODEL_NAME_TO_MODEL = {
     "mul": lambda: (MulModule(), MulModule.get_example_inputs()),
     "linear": lambda: (LinearModule(), LinearModule.get_example_inputs()),
@@ -103,4 +110,5 @@ def gen_torchvision_vit_model_and_inputs() -> Tuple[torch.nn.Module, Any]:
     "mv2": gen_mobilenet_v2_model_inputs,
     "mv3": gen_mobilenet_v3_model_inputs,
     "vit": gen_torchvision_vit_model_and_inputs,
+    "w2l": gen_wav2letter_model_and_inputs,
 }

examples/models/wav2letter/TARGETS

@@ -0,0 +1,14 @@
+load("@fbcode_macros//build_defs:python_library.bzl", "python_library")
+
+python_library(
+    name = "w2l_export",
+    srcs = [
+        "__init__.py",
+        "export.py",
+    ],
+    base_module = "executorch.examples.models.wav2letter",
+    deps = [
+        "//caffe2:torch",
+        "//pytorch/audio:torchaudio",
+    ],
+)

examples/models/wav2letter/__init__.py

@@ -0,0 +1,11 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from .export import Wav2LetterModel
+
+__all__ = [
+    Wav2LetterModel,
+]

examples/models/wav2letter/export.py

@@ -0,0 +1,30 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import logging
+
+import torch
+from torchaudio import models
+
+FORMAT = "[%(filename)s:%(lineno)s] %(message)s"
+logging.basicConfig(format=FORMAT)
+
+
+class Wav2LetterModel:
+    def __init__(self):
+        self.batch_size = 10
+        self.input_frames = 700
+        self.vocab_size = 4096
+
+    def get_model(self):
+        logging.info("loading wav2letter model")
+        wav2letter = models.Wav2Letter(num_classes=self.vocab_size)
+        logging.info("loaded wav2letter model")
+        return wav2letter
+
+    def get_example_inputs(self):
+        input_shape = (self.batch_size, 1, self.input_frames)
+        return (torch.randn(input_shape),)