Mark the subclass Transformer as deprecated.

Don't export the CompiledTransformer layer which may be deleted soon.

PiperOrigin-RevId: 478918916

official/nlp/modeling/layers/__init__.py

@@ -60,7 +60,8 @@ from official.nlp.modeling.layers.text_layers import BertTokenizer
 from official.nlp.modeling.layers.text_layers import FastWordpieceBertTokenizer
 from official.nlp.modeling.layers.text_layers import SentencepieceTokenizer
 from official.nlp.modeling.layers.tn_transformer_expand_condense import TNTransformerExpandCondense
-from official.nlp.modeling.layers.transformer import *
+from official.nlp.modeling.layers.transformer import Transformer
+from official.nlp.modeling.layers.transformer import TransformerDecoderBlock
 from official.nlp.modeling.layers.transformer_encoder_block import TransformerEncoderBlock
 from official.nlp.modeling.layers.transformer_scaffold import TransformerScaffold
 from official.nlp.modeling.layers.transformer_xl import TransformerXL

official/nlp/modeling/layers/transformer.py

@@ -15,6 +15,7 @@
 """Keras-based transformer block layer."""
 # pylint: disable=g-classes-have-attributes
 
+from absl import logging
 import gin
 import tensorflow as tf
 
@@ -32,6 +33,9 @@ class Transformer(transformer_encoder_block.TransformerEncoderBlock):
   This layer implements the Transformer from "Attention Is All You Need".
   (https://arxiv.org/abs/1706.03762).
 
+  **Warning: this layer is deprecated. Please don't use it. Use the
+  `TransformerEncoderBlock` layer instead.**
+
   Args:
     num_attention_heads: Number of attention heads.
     intermediate_size: Size of the intermediate layer.
@@ -98,6 +102,8 @@ class Transformer(transformer_encoder_block.TransformerEncoderBlock):
         inner_dropout=intermediate_dropout,
         attention_initializer=attention_initializer,
         **kwargs)
+    logging.warning("The `Transformer` layer is deprecated. Please directly "
+                    "use `TransformerEncoderBlock`.")
 
   def get_config(self):
     return {