diff --git a/neural_compressor/adaptor/tf_utils/graph_converter.py b/neural_compressor/adaptor/tf_utils/graph_converter.py index 72b0af00fd3..70d2b8af883 100644 --- a/neural_compressor/adaptor/tf_utils/graph_converter.py +++ b/neural_compressor/adaptor/tf_utils/graph_converter.py @@ -68,7 +68,7 @@ from .graph_util import GraphRewriterHelper as Helper -TF_SUPPORTED_MAX_VERSION = '2.9.1' +TF_SUPPORTED_MAX_VERSION = '2.10.0' TF_SUPPORTED_MIN_VERSION = '1.14.0' logger = logging.getLogger() @@ -247,7 +247,7 @@ def _check_tf_version(self): if version1_gte_version2(tf.version.VERSION, '2.9.0'): is_supported_version = True - if tf.version.VERSION == '1.15.0-up3': + if version1_eq_version2(tf.version.VERSION, '1.15.0-up3'): is_supported_version = True except Exception as e: diff --git a/neural_compressor/adaptor/tf_utils/graph_converter_without_calib.py b/neural_compressor/adaptor/tf_utils/graph_converter_without_calib.py index 99c987be76c..d1567dd41f2 100644 --- a/neural_compressor/adaptor/tf_utils/graph_converter_without_calib.py +++ b/neural_compressor/adaptor/tf_utils/graph_converter_without_calib.py @@ -44,9 +44,9 @@ from .graph_rewriter.int8.post_quantized_op_cse import PostCseOptimizer from .graph_rewriter.int8.meta_op_optimizer import MetaInfoChangingMemOpOptimizer from .graph_rewriter.int8.rnn_convert import QuantizedRNNConverter +from .util import version1_gte_version2,version1_gt_version2,version1_eq_version2 - -TF_SUPPORTED_MAX_VERSION = '2.9.1' +TF_SUPPORTED_MAX_VERSION = '2.10.0' TF_SUPPORTED_MIN_VERSION = '1.14.0' logger = logging.getLogger() @@ -107,16 +107,19 @@ def _check_tf_version(self): if IsMklEnabled() and (TF_SUPPORTED_MIN_VERSION <= tf.version.VERSION): is_supported_version = True - if tf.version.VERSION >= '2.6.0' and os.getenv('TF_ENABLE_ONEDNN_OPTS') == '1': + if version1_gte_version2(tf.version.VERSION, '2.6.0') and os.getenv('TF_ENABLE_ONEDNN_OPTS') == '1': + is_supported_version = True + + if version1_gte_version2(tf.version.VERSION, '2.9.0'): is_supported_version = True - if tf.version.VERSION >= '2.9.0': + if version1_eq_version2(tf.version.VERSION, '1.15.0-up3'): is_supported_version = True except Exception as e: raise ValueError(e) finally:# pragma: no cover - if tf.version.VERSION > TF_SUPPORTED_MAX_VERSION: + if version1_gt_version2(tf.version.VERSION, TF_SUPPORTED_MAX_VERSION): logger.warning( str('Please note the {} version of Intel® Optimizations for ' 'TensorFlow is not fully verified! ' @@ -124,12 +127,11 @@ def _check_tf_version(self): 'between {} and {} if meet problem.').format(tf.version.VERSION, TF_SUPPORTED_MIN_VERSION, TF_SUPPORTED_MAX_VERSION)) - if tf.version.VERSION == '2.5.0' and os.getenv('TF_ENABLE_MKL_NATIVE_FORMAT') != '0': + if version1_eq_version2(tf.version.VERSION, '2.5.0') and os.getenv('TF_ENABLE_MKL_NATIVE_FORMAT') != '0': logger.fatal("Please set environment variable TF_ENABLE_MKL_NATIVE_FORMAT=0 " "when TensorFlow 2.5.0 installed.") - if tf.version.VERSION >= '2.6.0' and tf.version.VERSION < '2.9.0' \ - and os.getenv('TF_ENABLE_ONEDNN_OPTS') != '1': + if version1_gte_version2(tf.version.VERSION, '2.6.0') and os.getenv('TF_ENABLE_ONEDNN_OPTS') != '1': logger.fatal("Please set environment variable TF_ENABLE_ONEDNN_OPTS=1 " "when TensorFlow >= 2.6.0 and < 2.9.0 installed.") diff --git a/neural_compressor/adaptor/tf_utils/graph_rewriter/generic/dummy_biasadd.py b/neural_compressor/adaptor/tf_utils/graph_rewriter/generic/dummy_biasadd.py index 19dc9149248..0f0b6a9d3e1 100644 --- a/neural_compressor/adaptor/tf_utils/graph_rewriter/generic/dummy_biasadd.py +++ b/neural_compressor/adaptor/tf_utils/graph_rewriter/generic/dummy_biasadd.py @@ -39,11 +39,8 @@ def do_transformation(self): valid_ops = ('BiasAdd', 'Add', 'AddV2', 'AddN') target_nodes = g.query_fusion_pattern_nodes([['MatMul', 'Conv2D'],]) for i in target_nodes: - # only apply this pass for tensorflow release 2.9.1 and lower version for - # old quantization API. + # only apply this pass for tensorflow old quantization API, pre_optimize does this check # use conv+dummy_biasadd+relu because TF do not support conv+relu now. - if version1_gt_version2(tf.version.VERSION, '2.9.1'): - continue if i[0] in self.outputs: continue next_node_names = graph_info[i[0]].outputs diff --git a/neural_compressor/adaptor/tf_utils/quantize_graph/quantize_graph_base.py b/neural_compressor/adaptor/tf_utils/quantize_graph/quantize_graph_base.py index f83fc9413d0..8f114c72412 100644 --- a/neural_compressor/adaptor/tf_utils/quantize_graph/quantize_graph_base.py +++ b/neural_compressor/adaptor/tf_utils/quantize_graph/quantize_graph_base.py @@ -627,7 +627,7 @@ def _add_quantize_down_nodes(self, # Add a RequantizationRange node for finding the min and max values. requant_range_node = helper.create_node( "RequantizationRangePerChannel" - if self.per_channel else "RequantizationRange", + if self.per_channel or original_node.op == 'DepthwiseConv2dNative' else "RequantizationRange", original_node.name + "_eightbit_requant_range", quantized_outputs) if self.per_channel: @@ -661,7 +661,8 @@ def _add_quantize_down_nodes(self, ] requantize_node = helper.create_node( - "RequantizePerChannel" if self.per_channel else "Requantize", + "RequantizePerChannel" if self.per_channel or original_node.op == 'DepthwiseConv2dNative' \ + else "Requantize", original_node.name + "_eightbit_requantize", quantized_outputs + min_max_inputs) if self.per_channel: diff --git a/test/adaptor/tensorflow_adaptor/test_tensorboard.py b/test/adaptor/tensorflow_adaptor/test_tensorboard.py index f9e3c15e88b..43c028670e6 100644 --- a/test/adaptor/tensorflow_adaptor/test_tensorboard.py +++ b/test/adaptor/tensorflow_adaptor/test_tensorboard.py @@ -10,6 +10,7 @@ from tensorflow.core.framework import node_def_pb2 from tensorflow.python.framework import tensor_util from tensorflow.python.framework import dtypes +from neural_compressor.adaptor.tf_utils.util import version1_gt_version2 tf.compat.v1.disable_eager_execution() def build_fake_yaml(): @@ -195,7 +196,8 @@ def tearDownClass(self): shutil.rmtree("saved", ignore_errors=True) shutil.rmtree("runs/", ignore_errors=True) - @unittest.skipIf(tf.version.VERSION > '2.5.0', " Skip test_bf16_fallback case for tf 2.6.0 and above.") + @unittest.skipIf(version1_gt_version2(tf.version.VERSION, '2.5.0'), \ + "Skip test_bf16_fallback case for tf 2.6.0 and above.") def test_run_basic_one_trial(self): from neural_compressor.experimental import Quantization, common