fix transformer amp

official/transformer/v2/misc.py

@@ -72,7 +72,8 @@ def define_transformer_flags():
       loss_scale=True,
       all_reduce_alg=True,
       enable_xla=True,
-      force_v2_in_keras_compile=True
+      force_v2_in_keras_compile=True,
+      fp16_implementation=True
   )
 
   # Additional performance flags
@@ -85,9 +86,6 @@ def define_transformer_flags():
            'convolutions and batch normalizations, and this flag allows to '
            'disable it.'
   )
-  flags.DEFINE_boolean(
-      name='automatic_mixed_precision', default=False,
-      help='Enable automatic mixed precision training via a graph rewrite.')
     
   flags_core.define_benchmark()
   flags_core.define_device(tpu=True)

official/transformer/v2/transformer_benchmark.py

@@ -330,6 +330,30 @@ class TransformerBigKerasAccuracy(TransformerBenchmark):
                                    bleu_min=28,
                                    bleu_max=29)
 
+  def benchmark_8_gpu_fp16_amp(self):
+    """Benchmark 8 gpu with dynamic batch and fp16 with automatic mixed precision.
+
+      Should converge to 28.4 BLEU (uncased). This has not be verified yet."
+    """
+    self._setup()
+    FLAGS.num_gpus = 8
+    FLAGS.dtype = 'fp16'
+    FLAGS.fp16_implementation = 'graph_rewrite'
+    FLAGS.data_dir = self.train_data_dir
+    FLAGS.vocab_file = self.vocab_file
+    # Sets values directly to avoid validation check.
+    FLAGS['bleu_source'].value = self.bleu_source
+    FLAGS['bleu_ref'].value = self.bleu_ref
+    FLAGS.param_set = 'big'
+    FLAGS.batch_size = 3072*8
+    FLAGS.train_steps = 20000 * 12
+    FLAGS.steps_between_evals = 20000
+    FLAGS.model_dir = self._get_model_dir('benchmark_8_gpu_fp16_amp')
+    self._run_and_report_benchmark(total_batch_size=FLAGS.batch_size,
+                                   log_steps=FLAGS.log_steps,
+                                   bleu_min=28,
+                                   bleu_max=29)
+    
   def benchmark_8_gpu_static_batch_fp16(self):
     """Benchmark 8 gpu with static batch and fp16.
 

official/transformer/v2/transformer_main.py

@@ -376,11 +376,12 @@ class TransformerTask(object):
       opt = tf.keras.mixed_precision.experimental.LossScaleOptimizer(
           opt, loss_scale=flags_core.get_loss_scale(self.flags_obj,
                                                     default_for_fp16="dynamic"))
-    if self.flags_obj.automatic_mixed_precision:
-        if params["dtype"] == tf.float16:
-            raise RuntimeError("Automatic mixed precision should not be called in conjunction with "
-                               "other types of mixed precision training. Set --dtype=fp32 instead.")
-        opt = tf.compat.v1.train.experimental.enable_mixed_precision_graph_rewrite(opt)
+    if self.flags_obj.fp16_implementation == "graph_rewrite":
+      # Note: when flags_obj.fp16_implementation == "graph_rewrite",
+      # dtype as determined by flags_core.get_tf_dtype(flags_obj) would be 'float32'
+      # which will ensure tf.keras.mixed_precision and tf.train.experimental.enable_mixed_precision_graph_rewrite
+      # does not double up.
+      opt = tf.train.experimental.enable_mixed_precision_graph_rewrite(opt)
     
     return opt
 

official/utils/flags/_performance.py

@@ -190,9 +190,9 @@ def define_performance(num_parallel_calls=True, inter_op=True, intra_op=True,
         return loss_scale > 0
 
     if fp16_implementation:
-      # Currently, this flag is only defined for the estimator resnet model.
+      # Currently, this flag is only defined for the estimator resnet and transformer models.
       flags.DEFINE_enum(
-          name="fp16_implementation", default="graph_rewrite",
+          name="fp16_implementation", default="casting",
           enum_values=("casting', 'graph_rewrite"),
           help=help_wrap(
               "When --dtype=fp16, how fp16 should be implemented. This has no "