[Mesh-TF] Add is_training as an arg to mtf.dropout

PiperOrigin-RevId: 361088273

tensor2tensor/models/mtf_image_transformer.py

@@ -243,8 +243,10 @@ def import_to_batch_by_length(x, name):
 def layer_prepostprocess_dropout(x, hparams):
   batch_dim = x.shape.dims[0]
   model_dim = x.shape.dims[-1]
+  mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+  is_training = mode == tf.estimator.ModeKeys.TRAIN
   return mtf.dropout(
-      x,
+      x, is_training,
       keep_prob=1.0 - hparams.layer_prepostprocess_dropout,
       noise_shape=mtf.Shape([batch_dim, model_dim]))
 
@@ -259,6 +261,8 @@ def local_attention1d_spatial_decoder(x, kv_dim, heads_dim,
   x = mtf.reshape(
       x, mtf.Shape([batch_dim, num_w_blocks_dim, blocks_w_dim, model_dim]))
   # [ self attention - ffn - residual + dropout] x n
+  mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+  is_training = mode == tf.estimator.ModeKeys.TRAIN
   for layer in range(hparams.num_decoder_layers):
     layer_name = "decoder_layer_%d" % layer
     with tf.variable_scope(layer_name):
@@ -268,6 +272,7 @@ def local_attention1d_spatial_decoder(x, kv_dim, heads_dim,
               mtf.layers.layer_norm(x, model_dim, name="layer_norm_att"),
               kv_dim,
               heads_dim,
+              is_training,
               memory_w_dim=blocks_w_dim,
               mask_right=True,
               name="self_att"), hparams)
@@ -276,6 +281,7 @@ def local_attention1d_spatial_decoder(x, kv_dim, heads_dim,
           mtf.layers.dense_relu_dense(
               mtf.layers.layer_norm(x, model_dim, name="layer_norm_ffn"),
               feedforward_dim,
+              is_training,
               hparams.dropout,
               dropout_broadcast_dims=[length_dim]), hparams)
 
@@ -305,6 +311,8 @@ def local_attention2d_spatial_decoder(x, kv_dim, heads_dim,
           batch_dim, num_h_blocks_dim, num_w_blocks_dim,
           blocks_h_dim, blocks_w_dim, model_dim
       ]))
+  mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+  is_training = mode == tf.estimator.ModeKeys.TRAIN
   # Image Transformer Decoder
   # [ self attention - ffn - residual + dropout] x n
   for layer in range(hparams.num_decoder_layers):
@@ -316,6 +324,7 @@ def local_attention2d_spatial_decoder(x, kv_dim, heads_dim,
               mtf.layers.layer_norm(x, model_dim, name="layer_norm_att"),
               kv_dim,
               heads_dim,
+              is_training,
               memory_h_dim=num_h_blocks_dim,
               memory_w_dim=num_w_blocks_dim,
               name="self_att"), hparams)
@@ -336,6 +345,8 @@ def local_attention1d_masked_decoder(x, kv_dim, heads_dim,
   """Image Transformer decoder with local1D masked layers."""
   print(x)
   _, length_dim, model_dim = x.shape.dims
+  mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+  is_training = mode == tf.estimator.ModeKeys.TRAIN
   for layer in range(hparams.num_decoder_layers):
     layer_name = "decoder_layer_%d" % layer
     with tf.variable_scope(layer_name):
@@ -347,6 +358,7 @@ def local_attention1d_masked_decoder(x, kv_dim, heads_dim,
               mtf.layers.layer_norm(x, model_dim, name="layer_norm_att"),
               kv_dim,
               heads_dim,
+              is_training,
               window_size=hparams.block_length,
               length_per_split=length_per_split,
               name="self_att"), hparams)

tensor2tensor/models/mtf_transformer.py

@@ -242,6 +242,8 @@ def _mtf_model_fn(self, features, mesh):
     hparams = self._hparams
     extra_losses = []
     targets = tf.to_int32(features["targets"])
+    mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+    is_training = mode == tf.estimator.ModeKeys.TRAIN
     if len(targets.get_shape()) > 2:
       tf.logging.info("targets = %s" % targets)
       targets = tf.squeeze(targets, [2, 3])
@@ -289,7 +291,7 @@ def pad_to_max_length(x):
 
     def layer_prepostprocess_dropout(x):
       return mtf.dropout(
-          x, keep_prob=1.0 - hparams.layer_prepostprocess_dropout,
+          x, is_training, keep_prob=1.0 - hparams.layer_prepostprocess_dropout,
           noise_shape=mtf.Shape(self.batch_dims + [self.model_dim]))
 
     (inputs_embedding_var,
@@ -426,10 +428,11 @@ def _feedforward_layer(self, x, layer_type, losses=None):
       ValueError: if hparams make no sense
     """
     hparams = self._hparams
-
+    mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+    is_training = mode == tf.estimator.ModeKeys.TRAIN
     if layer_type == "drd":
       return mtf.layers.dense_relu_dense(
-          x, self.feedforward_dim, dropout=hparams.relu_dropout,
+          x, self.feedforward_dim, is_training, dropout=hparams.relu_dropout,
           dropout_broadcast_dims=[self.length_dim],
           master_dtype=self.master_dtype,
           slice_dtype=self.slice_dtype)
@@ -493,11 +496,13 @@ def _layer_stack(self,
     """
     hparams = self._hparams
     is_incremental = (step_num is not None)
+    mode = getattr(hparams, "mode", tf.estimator.ModeKeys.TRAIN)
+    is_training = mode == tf.estimator.ModeKeys.TRAIN
     def layer_prepostprocess_dropout(x):
       if is_incremental:
         return x
       return mtf.dropout(
-          x, keep_prob=1.0 - hparams.layer_prepostprocess_dropout,
+          x, is_training, keep_prob=1.0 - hparams.layer_prepostprocess_dropout,
           noise_shape=mtf.Shape(self.batch_dims + [self.model_dim]))
     num_layers = len(layers)
     num_layer_norms = num_layers + 1
@@ -540,6 +545,7 @@ def normalize(x):
                 mtf.layers.multihead_attention(
                     normalize(x), None,
                     self_attention_mask, self.kv_dim, self.heads_dim,
+                    is_training,
                     dropout=hparams.attention_dropout,
                     dropout_broadcast_dims=[self.length_dim],
                     master_dtype=self.master_dtype,
@@ -560,6 +566,7 @@ def normalize(x):
                 mtf.layers.multihead_attention(
                     normalize(x), encoder_output,
                     encdec_attention_mask, self.kv_dim, self.heads_dim,
+                    is_training,
                     dropout=hparams.attention_dropout,
                     dropout_broadcast_dims=[self.length_dim],
                     master_dtype=self.master_dtype,
@@ -582,7 +589,7 @@ def normalize(x):
             x += layer_prepostprocess_dropout(
                 mtf.layers.masked_local_attention_1d(
                     normalize(x),
-                    self.kv_dim, self.heads_dim,
+                    self.kv_dim, self.heads_dim, is_training,
                     window_size=hparams.local_attention_window_size,
                     master_dtype=self.master_dtype,
                     slice_dtype=self.slice_dtype,
@@ -601,6 +608,7 @@ def normalize(x):
                     compression_factor=hparams.compression_factor,
                     kv_channels=self.kv_dim,
                     heads=self.heads_dim,
+                    is_training=is_training,
                     dropout=hparams.attention_dropout,
                     dropout_broadcast_dims=[self.length_dim],
                     master_dtype=self.master_dtype,