Skip to content

Commit a1c6041

Browse files
JackHuntgeorgepaw
authored andcommitted
Keras EffectiveTransformer layer.
Summary: IPU implementation of https://github.com/bytedance/effective_transformer TF2.4 Only Test Plan: Inference and training tests. Unpadding test. Arg tests. Reviewers: #tensorflow, simonl, #framework_ip_review_-_any_oss_or_third-party_code_use_has_been_approved, samuelm, nicolasc, georgep Reviewed By: #tensorflow, #framework_ip_review_-_any_oss_or_third-party_code_use_has_been_approved, samuelm, nicolasc, georgep Subscribers: jamiep, markf, mrinal, georgep Maniphest Tasks: T35014 Differential Revision: https://phabricator.sourcevertex.net/D43654
1 parent dc90fd9 commit a1c6041

File tree

5 files changed

+1247
-1
lines changed

5 files changed

+1247
-1
lines changed

tensorflow/compiler/plugin/poplar/docs/api.rst

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -167,7 +167,7 @@ Keras layers
167167

168168

169169
.. automodule:: tensorflow.python.ipu.keras.layers
170-
:members: Dropout, Embedding, GroupNormalization, InstanceNormalization, LayerNormalization, PopnnGRU, PopnnLSTM, SerialDense, CTCInferenceLayer, CTCPredictionsLayer, RecomputationCheckpoint
170+
:members: Dropout, EffectiveTransformer, Embedding, GroupNormalization, InstanceNormalization, LayerNormalization, PopnnGRU, PopnnLSTM, SerialDense, CTCInferenceLayer, CTCPredictionsLayer, RecomputationCheckpoint
171171
:imported-members: Dropout, Embedding, GroupNormalization, InstanceNormalization, LayerNormalization, PopnnGRU, PopnnLSTM, SerialDense, CTCInferenceLayer, CTCPredictionsLayer, RecomputationCheckpoint
172172

173173
Keras losses

tensorflow/python/ipu/BUILD

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -126,6 +126,7 @@ py_library(
126126
"keras/layers/assume_equal_across_replicas.py",
127127
"keras/layers/ctc.py",
128128
"keras/layers/dropout.py",
129+
"keras/layers/effective_transformer.py",
129130
"keras/layers/embedding_lookup.py",
130131
"keras/layers/ipu_layer.py",
131132
"keras/layers/math.py",
@@ -538,6 +539,18 @@ tf_py_test(
538539
],
539540
)
540541

542+
tf_py_test(
543+
name = "keras_effective_transformer_test",
544+
size = "large",
545+
srcs = ["tests/keras/keras_effective_transformer_test.py"],
546+
shard_count = 12,
547+
deps = [
548+
"//tensorflow/python:framework_test_lib",
549+
"//tensorflow/python/ipu:ipu_lib",
550+
"//third_party/py/numpy",
551+
],
552+
)
553+
541554
tf_py_test(
542555
name = "keras_mnist_test",
543556
size = "large",
@@ -1963,6 +1976,7 @@ test_suite(
19631976
"keras_asynchronous_callbacks_test",
19641977
"keras_ctc_test",
19651978
"keras_dropout_test",
1979+
"keras_effective_transformer_test",
19661980
"keras_embedding_lookup_test",
19671981
"keras_execution_parameters_test",
19681982
"keras_functional_model_test",

tensorflow/python/ipu/keras/layers/__init__.py

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -21,6 +21,7 @@
2121
from tensorflow.python.ipu.keras.layers.ctc import CTCInferenceLayer
2222
from tensorflow.python.ipu.keras.layers.ctc import CTCPredictionsLayer
2323
from tensorflow.python.ipu.keras.layers.dropout import Dropout
24+
from tensorflow.python.ipu.keras.layers.effective_transformer import EffectiveTransformer
2425
from tensorflow.python.ipu.keras.layers.embedding_lookup import Embedding
2526
from tensorflow.python.ipu.keras.layers.normalization import GroupNorm, InstanceNorm, LayerNorm
2627
from tensorflow.python.ipu.keras.layers.normalization import GroupNormalization, InstanceNormalization, LayerNormalization

0 commit comments

Comments
 (0)