Merge pull request #7485 from reedwm:mixed_float16_transformer

PiperOrigin-RevId: 265483790

official/transformer/v2/embedding_layer.py

@@ -24,24 +24,14 @@ import tensorflow as tf
 class EmbeddingSharedWeights(tf.keras.layers.Layer):
   """Calculates input embeddings and pre-softmax linear with shared weights."""
 
-  def __init__(self, vocab_size, hidden_size, dtype=None):
+  def __init__(self, vocab_size, hidden_size):
     """Specify characteristic parameters of embedding layer.
 
     Args:
       vocab_size: Number of tokens in the embedding. (Typically ~32,000)
       hidden_size: Dimensionality of the embedding. (Typically 512 or 1024)
-      dtype: The dtype of the layer: float16 or float32.
     """
-    if dtype == tf.float16:
-      # We cannot rely on the global policy of "infer_with_float32_vars", as
-      # this layer is called on both int64 inputs and floating-point inputs.
-      # If "infer_with_float32_vars" is used, the dtype will be inferred to be
-      # int64, which means floating-point inputs would not be casted.
-      # TODO(b/138859351): Remove this logic once we stop using the deprecated
-      # "infer_with_float32_vars" policy
-      dtype = tf.keras.mixed_precision.experimental.Policy(
-          "float16_with_float32_vars")
-    super(EmbeddingSharedWeights, self).__init__(dtype=dtype)
+    super(EmbeddingSharedWeights, self).__init__()
     self.vocab_size = vocab_size
     self.hidden_size = hidden_size
 
@@ -53,7 +43,6 @@ class EmbeddingSharedWeights(tf.keras.layers.Layer):
       self.shared_weights = self.add_weight(
           "weights",
           shape=[self.vocab_size, self.hidden_size],
-          dtype="float32",
           initializer=tf.random_normal_initializer(
               mean=0., stddev=self.hidden_size**-0.5))
     super(EmbeddingSharedWeights, self).build(input_shape)

official/transformer/v2/transformer.py

@@ -49,8 +49,10 @@ def create_model(params, is_train):
       label_smoothing = params["label_smoothing"]
       if params["enable_metrics_in_training"]:
         logits = metrics.MetricLayer(vocab_size)([logits, targets])
-      logits = tf.keras.layers.Lambda(lambda x: x, name="logits")(logits)
+      logits = tf.keras.layers.Lambda(lambda x: x, name="logits",
+                                      dtype=tf.float32)(logits)
       model = tf.keras.Model([inputs, targets], logits)
+      # TODO(reedwm): Can we do this loss in float16 instead of float32?
       loss = metrics.transformer_loss(
           logits, targets, label_smoothing, vocab_size)
       model.add_loss(loss)
@@ -85,7 +87,7 @@ class Transformer(tf.keras.Model):
     super(Transformer, self).__init__(name=name)
     self.params = params
     self.embedding_softmax_layer = embedding_layer.EmbeddingSharedWeights(
-        params["vocab_size"], params["hidden_size"], dtype=params["dtype"])
+        params["vocab_size"], params["hidden_size"])
     self.encoder_stack = EncoderStack(params)
     self.decoder_stack = DecoderStack(params)
 

official/transformer/v2/transformer_layers_test.py

+# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+#
+# 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.
+# ==============================================================================
 """Tests for layers in Transformer."""
 
 from __future__ import absolute_import
@@ -79,4 +93,5 @@ class TransformerLayersTest(tf.test.TestCase):
 
 
 if __name__ == "__main__":
+  tf.compat.v1.enable_v2_behavior()
   tf.test.main()

official/transformer/v2/transformer_main.py

@@ -168,8 +168,10 @@ class TransformerTask(object):
       # like this. What if multiple instances of TransformerTask are created?
       # We should have a better way in the tf.keras.mixed_precision API of doing
       # this.
+      loss_scale = flags_core.get_loss_scale(flags_obj,
+                                             default_for_fp16="dynamic")
       policy = tf.keras.mixed_precision.experimental.Policy(
-          "infer_float32_vars")
+          "mixed_float16", loss_scale=loss_scale)
       tf.keras.mixed_precision.experimental.set_policy(policy)
 
     self.distribution_strategy = distribution_utils.get_distribution_strategy(
@@ -417,10 +419,6 @@ class TransformerTask(object):
         params["optimizer_adam_beta1"],
         params["optimizer_adam_beta2"],
         epsilon=params["optimizer_adam_epsilon"])
-    if params["dtype"] == tf.float16:
-      opt = tf.keras.mixed_precision.experimental.LossScaleOptimizer(
-          opt, loss_scale=flags_core.get_loss_scale(self.flags_obj,
-                                                    default_for_fp16="dynamic"))
     return opt
 
 

official/transformer/v2/transformer_main_test.py

@@ -184,4 +184,5 @@ class TransformerTaskTest(tf.test.TestCase):
 
 
 if __name__ == '__main__':
+  tf.compat.v1.enable_v2_behavior()
   tf.test.main()

official/transformer/v2/transformer_test.py

@@ -65,4 +65,5 @@ class TransformerV2Test(tf.test.TestCase):
 
 
 if __name__ == "__main__":
+  tf.compat.v1.enable_v2_behavior()
   tf.test.main()

research/lstm_object_detection/export_tflite_lstd_graph.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
+
 r"""Exports an LSTM detection model to use with tf-lite.
 
 Outputs file:
@@ -85,9 +86,8 @@ python lstm_object_detection/export_tflite_lstd_graph.py \
 """
 
 import tensorflow as tf
-
-from lstm_object_detection import export_tflite_lstd_graph_lib
 from lstm_object_detection.utils import config_util
+from lstm_object_detection import export_tflite_lstd_graph_lib
 
 flags = tf.app.flags
 flags.DEFINE_string('output_directory', None, 'Path to write outputs.')
@@ -125,13 +125,9 @@ def main(argv):
     FLAGS.pipeline_config_path)
 
   export_tflite_lstd_graph_lib.export_tflite_graph(
-      pipeline_config,
-      FLAGS.trained_checkpoint_prefix,
-      FLAGS.output_directory,
-      FLAGS.add_postprocessing_op,
-      FLAGS.max_detections,
-      FLAGS.max_classes_per_detection,
-      use_regular_nms=FLAGS.use_regular_nms)
+      pipeline_config, FLAGS.trained_checkpoint_prefix, FLAGS.output_directory,
+      FLAGS.add_postprocessing_op, FLAGS.max_detections,
+      FLAGS.max_classes_per_detection, use_regular_nms=FLAGS.use_regular_nms)
 
 
 if __name__ == '__main__':

research/lstm_object_detection/export_tflite_lstd_graph_lib.py

@@ -12,26 +12,26 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
+
 r"""Exports detection models to use with tf-lite.
 
 See export_tflite_lstd_graph.py for usage.
 """
 import os
 import tempfile
-
 import numpy as np
 import tensorflow as tf
-
 from tensorflow.core.framework import attr_value_pb2
 from tensorflow.core.framework import types_pb2
 from tensorflow.core.protobuf import saver_pb2
 from tensorflow.tools.graph_transforms import TransformGraph
-from lstm_object_detection import model_builder
 from object_detection import exporter
 from object_detection.builders import graph_rewriter_builder
 from object_detection.builders import post_processing_builder
 from object_detection.core import box_list
 
+from lstm_object_detection import model_builder
+
 _DEFAULT_NUM_CHANNELS = 3
 _DEFAULT_NUM_COORD_BOX = 4
 
@@ -87,8 +87,8 @@ def append_postprocessing_op(frozen_graph_def,
       centersize boxes
     detections_per_class: In regular NonMaxSuppression, number of anchors used
     for NonMaxSuppression per class
-    use_regular_nms: Flag to set postprocessing op to use Regular NMS instead of
-      Fast NMS.
+    use_regular_nms: Flag to set postprocessing op to use Regular NMS instead
+      of Fast NMS.
 
   Returns:
     transformed_graph_def: Frozen GraphDef with postprocessing custom op
@@ -165,9 +165,9 @@ def export_tflite_graph(pipeline_config,
   is written to output_dir/tflite_graph.pb.
 
   Args:
-    pipeline_config: Dictionary of configuration objects. Keys are `model`,
-      `train_config`, `train_input_config`, `eval_config`, `eval_input_config`,
-      `lstm_model`. Value are the corresponding config objects.
+    pipeline_config: Dictionary of configuration objects. Keys are `model`, `train_config`,
+      `train_input_config`, `eval_config`, `eval_input_config`, `lstm_model`.
+      Value are the corresponding config objects.
     trained_checkpoint_prefix: a file prefix for the checkpoint containing the
       trained parameters of the SSD model.
     output_dir: A directory to write the tflite graph and anchor file to.
@@ -177,8 +177,8 @@ def export_tflite_graph(pipeline_config,
     max_classes_per_detection: Number of classes to display per detection
     detections_per_class: In regular NonMaxSuppression, number of anchors used
     for NonMaxSuppression per class
-    use_regular_nms: Flag to set postprocessing op to use Regular NMS instead of
-      Fast NMS.
+    use_regular_nms: Flag to set postprocessing op to use Regular NMS instead
+      of Fast NMS.
     binary_graph_name: Name of the exported graph file in binary format.
     txt_graph_name: Name of the exported graph file in text format.
 
@@ -197,10 +197,12 @@ def export_tflite_graph(pipeline_config,
 
   num_classes = model_config.ssd.num_classes
   nms_score_threshold = {
-      model_config.ssd.post_processing.batch_non_max_suppression.score_threshold
+      model_config.ssd.post_processing.batch_non_max_suppression.
+      score_threshold
   }
   nms_iou_threshold = {
-      model_config.ssd.post_processing.batch_non_max_suppression.iou_threshold
+      model_config.ssd.post_processing.batch_non_max_suppression.
+      iou_threshold
   }
   scale_values = {}
   scale_values['y_scale'] = {
@@ -224,7 +226,7 @@ def export_tflite_graph(pipeline_config,
     width = image_resizer_config.fixed_shape_resizer.width
     if image_resizer_config.fixed_shape_resizer.convert_to_grayscale:
       num_channels = 1
-
+    #TODO(richardbrks) figure out how to make with a None defined batch size
     shape = [lstm_config.eval_unroll_length, height, width, num_channels]
   else:
     raise ValueError(
@@ -235,8 +237,8 @@ def export_tflite_graph(pipeline_config,
   video_tensor = tf.placeholder(
     tf.float32, shape=shape, name='input_video_tensor')
 
-  detection_model = model_builder.build(
-      model_config, lstm_config, is_training=False)
+  detection_model = model_builder.build(model_config, lstm_config,
+    is_training=False)
   preprocessed_video, true_image_shapes = detection_model.preprocess(
     tf.to_float(video_tensor))
   predicted_tensors = detection_model.predict(preprocessed_video,
@@ -309,7 +311,7 @@ def export_tflite_graph(pipeline_config,
       initializer_nodes='')
 
   # Add new operation to do post processing in a custom op (TF Lite only)
-
+  #(richardbrks) Do use this or detection_model.postprocess?
   if add_postprocessing_op:
     transformed_graph_def = append_postprocessing_op(
         frozen_graph_def, max_detections, max_classes_per_detection,

research/lstm_object_detection/export_tflite_lstd_model.py

@@ -13,8 +13,6 @@
 # limitations under the License.
 # ==============================================================================
 
-"""Export a LSTD model in tflite format."""
-
 import os
 from absl import flags
 import tensorflow as tf
@@ -51,14 +49,13 @@ def main(_):
     }
 
     converter = tf.lite.TFLiteConverter.from_frozen_graph(
-      FLAGS.frozen_graph_path,
-      input_arrays,
-      output_arrays,
-      input_shapes=input_shapes)
+        FLAGS.frozen_graph_path, input_arrays, output_arrays,
+        input_shapes=input_shapes
+    )
     converter.allow_custom_ops = True
     tflite_model = converter.convert()
     ofilename = os.path.join(FLAGS.export_path)
-  open(ofilename, 'wb').write(tflite_model)
+    open(ofilename, "wb").write(tflite_model)
 
 
 if __name__ == '__main__':

research/lstm_object_detection/g3doc/exporting_models.md

 # Exporting a tflite model from a checkpoint
 
-Starting from a trained model checkpoint, creating a tflite model requires 2
-steps:
+Starting from a trained model checkpoint, creating a tflite model requires 2 steps:
 
 * exporting a tflite frozen graph from a checkpoint
 * exporting a tflite model from a frozen graph
 
+
 ## Exporting a tflite frozen graph from a checkpoint
 
 With a candidate checkpoint to export, run the following command from
@@ -23,12 +23,12 @@ python lstm_object_detection/export_tflite_lstd_graph.py \
 	--add_preprocessing_op
 ```
 
-After export, you should see the directory ${EXPORT_DIR} containing the
-following files:
+After export, you should see the directory ${EXPORT_DIR} containing the following files:
 
 * `tflite_graph.pb`
 * `tflite_graph.pbtxt` 
 
+
 ## Exporting a tflite model from a frozen graph
 
 We then take the exported tflite-compatable tflite model, and convert it to a

research/lstm_object_detection/test_tflite_model.py

@@ -13,9 +13,6 @@
 # limitations under the License.
 # ==============================================================================
 
-"""Test a tflite model using random input data."""
-
-from __future__ import print_function
 from absl import flags
 import numpy as np
 import tensorflow as tf
@@ -34,9 +31,9 @@ def main(_):
 
     # Get input and output tensors.
     input_details = interpreter.get_input_details()
-  print('input_details:', input_details)
+    print 'input_details:', input_details
     output_details = interpreter.get_output_details()
-  print('output_details:', output_details)
+    print 'output_details:', output_details
 
     # Test model on random input data.
     input_shape = input_details[0]['shape']
@@ -46,7 +43,7 @@ def main(_):
 
     interpreter.invoke()
     output_data = interpreter.get_tensor(output_details[0]['index'])
-  print(output_data)
+    print output_data
 
 
 if __name__ == '__main__':

research/lstm_object_detection/tflite/BUILD

@@ -59,19 +59,12 @@ cc_library(
     name = "mobile_lstd_tflite_client",
     srcs = ["mobile_lstd_tflite_client.cc"],
     hdrs = ["mobile_lstd_tflite_client.h"],
-    defines = select({
-        "//conditions:default": [],
-        "enable_edgetpu": ["ENABLE_EDGETPU"],
-    }),
     deps = [
         ":mobile_ssd_client",
         ":mobile_ssd_tflite_client",
-        "@com_google_glog//:glog",
         "@com_google_absl//absl/base:core_headers",
+        "@com_google_glog//:glog",
         "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
-    ] + select({
-        "//conditions:default": [],
-        "enable_edgetpu": ["@libedgetpu//libedgetpu:header"],
-    }),
+    ],
     alwayslink = 1,
 )

research/lstm_object_detection/tflite/WORKSPACE

@@ -90,6 +90,12 @@ http_archive(
     sha256 = "79d102c61e2a479a0b7e5fc167bcfaa4832a0c6aad4a75fa7da0480564931bcc",
 )
 
+#
+# http_archive(
+#     name = "com_google_protobuf",
+#     strip_prefix = "protobuf-master",
+#     urls = ["https://github.com/protocolbuffers/protobuf/archive/master.zip"],
+# )
 
 # Needed by TensorFlow
 http_archive(

research/lstm_object_detection/tflite/mobile_lstd_tflite_client.cc

@@ -66,11 +66,6 @@ bool MobileLSTDTfLiteClient::InitializeInterpreter(
     interpreter_->UseNNAPI(false);
   }
 
-#ifdef ENABLE_EDGETPU
-  interpreter_->SetExternalContext(kTfLiteEdgeTpuContext,
-                                   edge_tpu_context_.get());
-#endif
-
   // Inputs are: normalized_input_image_tensor, raw_inputs/init_lstm_c,
   // raw_inputs/init_lstm_h
   if (interpreter_->inputs().size() != 3) {

research/lstm_object_detection/tflite/mobile_ssd_tflite_client.h

@@ -76,10 +76,6 @@ class MobileSSDTfLiteClient : public MobileSSDClient {
   std::unique_ptr<::tflite::MutableOpResolver> resolver_;
   std::unique_ptr<::tflite::Interpreter> interpreter_;
 
-#ifdef ENABLE_EDGETPU
-  std::unique_ptr<edgetpu::EdgeTpuContext> edge_tpu_context_;
-#endif
-
  private:
   // MobileSSDTfLiteClient is neither copyable nor movable.
   MobileSSDTfLiteClient(const MobileSSDTfLiteClient&) = delete;
@@ -107,6 +103,10 @@ class MobileSSDTfLiteClient : public MobileSSDClient {
   bool FloatInference(const uint8_t* input_data);
   bool QuantizedInference(const uint8_t* input_data);
   void GetOutputBoxesAndScoreTensorsFromUInt8();
+
+#ifdef ENABLE_EDGETPU
+  std::unique_ptr<edgetpu::EdgeTpuContext> edge_tpu_context_;
+#endif
 };
 
 }  // namespace tflite