Merge branch 'fix_cmake_package_TF' into 'master'

tnt: optimizers: use `tnt.optimizers.Optimizer` naming scheme

See merge request carpenamarie/hpdlf!215

cmake/FindTensorflow.cmake

@@ -90,7 +90,6 @@ include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(Tensorflow
                                   REQUIRED_VARS Tensorflow_LIBRARY
                                                 Tensorflow_INCLUDE_DIR
-                                                Tensorflow_CXX11_ABI_FLAG
                                   VERSION_VAR Tensorflow_VERSION)
 
 mark_as_advanced(Tensorflow_INCLUDE_DIR

src/mypy.ini

@@ -0,0 +1,50 @@
+[mypy]
+disallow_untyped_calls = False
+disallow_untyped_defs = False
+disallow_incomplete_defs = True
+disallow_subclassing_any = False
+check_untyped_defs= True
+
+warn_unused_ignores = True
+warn_no_return = True
+warn_return_any = True
+warn_unreachable = True
+warn_unused_configs = True
+
+show_none_errors = True
+show_error_codes = True
+
+allow_untyped_globals = False
+allow_redefinition = False
+
+#strict = True
+#implicit_reexport = True # prevent Module has no attribute errors
+
+linecount_report = coverage
+linecoverage_report = coverage
+lineprecision_report = coverage
+
+exclude = (?x)(
+    /examples
+    | /gpi_comm_lib
+    | /\.mypy
+    | tarantella/keras/utilities\.py
+  )
+
+[mypy-tensorflow.*]
+ignore_missing_imports = True
+
+[mypy-networkx.*]
+ignore_missing_imports = True
+
+[mypy-pygpi]
+ignore_missing_imports = True
+follow_imports=skip
+
+[mypy-tnt_tfops]
+ignore_missing_imports = True
+follow_imports=skip
+
+[mypy-GPICommLib]
+ignore_missing_imports = True
+follow_imports=skip

src/tarantella/__init__.py

@@ -55,5 +55,3 @@ def is_group_master_rank(group: pygpi.Group) -> bool:
 from tarantella.collectives.TensorAllgatherer import TensorAllgatherer
 
 import tarantella.optimizers as optimizers
-from tarantella.optimizers.synchronous_distributed_optimizer import Optimizer
-

src/tarantella/keras/model.py

@@ -18,8 +18,7 @@ def __call__(cls, *args, **kwargs):
     return obj
 
   def _create_tnt_model(cls, model: tf.keras.Model,
-                        parallel_strategy: tnt.ParallelStrategy = tnt.ParallelStrategy.ALL if TF_DEFAULT_PIPELINING_FLAG \
-                                                                                           else tnt.ParallelStrategy.DATA,
+                        parallel_strategy: tnt.ParallelStrategy = tnt.ParallelStrategy.DATA,
                         num_pipeline_stages: int = 1):
     replica_group = tnt.Group()
 

src/tarantella/keras/models.py

@@ -22,8 +22,8 @@ def load_model(filepath, compile = True, **kwargs):
   tnt_model = tnt.Model(keras_model, parallel_strategy = tnt.ParallelStrategy.DATA)
   if compile:
     try:
-      tnt_optimizer = tnt.distributed_optimizers.SynchDistributedOptimizer(keras_model.optimizer,
-                                                                           group = tnt_model.group)
+      tnt_optimizer = tnt.optimizers.Optimizer(keras_model.optimizer,
+                                               group = tnt_model.group)
       tnt_model.dist_optimizer = tnt_optimizer
       tnt_model._set_internal_optimizer(tnt_model.dist_optimizer)
       tnt_model.compiled = True

src/tarantella/optimizers/__init__.py

@@ -1 +1 @@
-
+from tarantella.optimizers.synchronous_distributed_optimizer import Optimizer

src/tarantella/optimizers/synchronous_distributed_optimizer.py

@@ -22,7 +22,7 @@ def __init__(self, keras_optimizer: tf.keras.optimizers.Optimizer,
                        name: str = None,
                        group: tnt.Group = None):
       self.keras_optimizer = keras_optimizer
-      logger.debug(f"[SynchDistributedOptimizer] Initializing generic tnt.Optimizer of type={type(keras_optimizer)}")
+      logger.debug(f"[SynchDistributedOptimizer] Initializing generic tnt.optimizers.Optimizer of type={type(keras_optimizer)}")
       _construct_from_keras_object(self, keras_optimizer)
 
       if name is None:

src/tarantella/strategy/data_parallel/data_parallel_model.py

@@ -46,7 +46,7 @@ def compile(self,
     elif isinstance(optimizer, str):
       config = {'class_name': optimizer, 'config': {}}
       optimizer = tf.keras.optimizers.deserialize(config)
-    self.dist_optimizer = tnt.Optimizer(optimizer, group = self.group)
+    self.dist_optimizer = tnt.optimizers.Optimizer(optimizer, group = self.group)
 
     kwargs = self._preprocess_compile_kwargs(kwargs)
     return self.model.compile(optimizer = self.dist_optimizer,

test/python/model_api.py

@@ -1,10 +1,7 @@
 from models import mnist_models as mnist
-from tarantella.strategy.parallel_strategy import ParallelStrategy
-import training_runner as base_runner
 import utilities as util
 import tarantella as tnt
 
-import numpy as np
 import tensorflow as tf
 from tensorflow import keras
 from tensorflow.keras import layers
@@ -168,5 +165,4 @@ def test_optimizer_with_name(self, optimizer_name, optimizer_type):
                       loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
                       metrics=['accuracy'])
     tnt_optimizer = tnt_model.dist_optimizer
-    assert isinstance(tnt_optimizer, tnt.distributed_optimizers.SynchDistributedOptimizer)
-    assert isinstance(tnt_optimizer.underlying_optimizer, optimizer_type)
+    assert isinstance(tnt_optimizer, optimizer_type)