From e374d97b6dea8442e44b8d9e308b02e88e62dcf8 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 02:52:26 +0000 Subject: [PATCH 01/13] update TensorFlow/built-in/recommendation/MMoE_Transformer_ID0390_for_TensorFlow/train/main.py. --- .../MMoE_Transformer_ID0390_for_TensorFlow/train/main.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/recommendation/MMoE_Transformer_ID0390_for_TensorFlow/train/main.py b/TensorFlow/built-in/recommendation/MMoE_Transformer_ID0390_for_TensorFlow/train/main.py index ae177ccdf..c15c1fe2b 100644 --- a/TensorFlow/built-in/recommendation/MMoE_Transformer_ID0390_for_TensorFlow/train/main.py +++ b/TensorFlow/built-in/recommendation/MMoE_Transformer_ID0390_for_TensorFlow/train/main.py @@ -255,6 +255,7 @@ if __name__ == '__main__': custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes('allow_mix_precision') custom_op.parameter_map["dynamic_input"].b = True custom_op.parameter_map["dynamic_graph_execute_mode"].s = tf.compat.as_bytes("lazy_recompile") + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") if iterations_per_loop_mode: custom_op.parameter_map["iterations_per_loop"].i = 10 sess_config.graph_options.rewrite_options.remapping = RewriterConfig.OFF -- Gitee From 3a2f922d155ddea33293a3c80eadf48407a01e04 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:05:08 +0000 Subject: [PATCH 02/13] update TensorFlow/contrib/cv/NOISE2NOISE_ID0800_for_TensorFlow/dnnlib/tflib/tfutil.py. --- .../NOISE2NOISE_ID0800_for_TensorFlow/dnnlib/tflib/tfutil.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/TensorFlow/contrib/cv/NOISE2NOISE_ID0800_for_TensorFlow/dnnlib/tflib/tfutil.py b/TensorFlow/contrib/cv/NOISE2NOISE_ID0800_for_TensorFlow/dnnlib/tflib/tfutil.py index 9f0c8cf74..63d2fef8a 100644 --- a/TensorFlow/contrib/cv/NOISE2NOISE_ID0800_for_TensorFlow/dnnlib/tflib/tfutil.py +++ b/TensorFlow/contrib/cv/NOISE2NOISE_ID0800_for_TensorFlow/dnnlib/tflib/tfutil.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. +# Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # This work is licensed under the Creative Commons Attribution-NonCommercial # 4.0 International License. To view a copy of this license, visit @@ -96,6 +96,7 @@ def create_session(config_dict: dict = None, force_as_default: bool = False, gra custom_op.parameter_map["mix_compile_mode"].b = config_dict["mix_compile_mode"] custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes(config_dict["precision_mode"]) custom_op.parameter_map["op_select_implmode"].s = tf.compat.as_bytes(config_dict["op_select_implmode"]) + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") # 开启Allreduce和前后向并行执行 custom_op.parameter_map["hcom_parallel"].b = config_dict["hcom_parallel"] -- Gitee From 39ae6f59623553eac92ecb04b8508481347342a8 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:08:40 +0000 Subject: [PATCH 03/13] update TensorFlow/contrib/cv/VDSR_ID2114_for_TensorFlow/VDSR.py. --- TensorFlow/contrib/cv/VDSR_ID2114_for_TensorFlow/VDSR.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/contrib/cv/VDSR_ID2114_for_TensorFlow/VDSR.py b/TensorFlow/contrib/cv/VDSR_ID2114_for_TensorFlow/VDSR.py index 5c44dd5d0..0e32f291c 100644 --- a/TensorFlow/contrib/cv/VDSR_ID2114_for_TensorFlow/VDSR.py +++ b/TensorFlow/contrib/cv/VDSR_ID2114_for_TensorFlow/VDSR.py @@ -185,6 +185,7 @@ if __name__ == '__main__': custom_op = config.graph_options.rewrite_options.custom_optimizers.add() custom_op.name = "NpuOptimizer" custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes("allow_mix_precision") + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") config.graph_options.rewrite_options.remapping = RewriterConfig.OFF # 必须显式关闭remap # custom_op.parameter_map["use_off_line"].b = True # config.graph_options.rewrite_options.memory_optimization = RewriterConfig.OFF -- Gitee From 7602e42447fb0832800333ed3a9b4c9dd9045495 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:30:23 +0000 Subject: [PATCH 04/13] image_classification/VGG16_ID0068_for_TensorFlow/vgg16/create_session.py. --- .../VGG16_ID0068_for_TensorFlow/vgg16/create_session.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/cv/image_classification/VGG16_ID0068_for_TensorFlow/vgg16/create_session.py b/TensorFlow/built-in/cv/image_classification/VGG16_ID0068_for_TensorFlow/vgg16/create_session.py index 3057d192b..e30135456 100644 --- a/TensorFlow/built-in/cv/image_classification/VGG16_ID0068_for_TensorFlow/vgg16/create_session.py +++ b/TensorFlow/built-in/cv/image_classification/VGG16_ID0068_for_TensorFlow/vgg16/create_session.py @@ -37,6 +37,7 @@ class CreateSession(): custom_op.name = "NpuOptimizer" custom_op.parameter_map["use_off_line"].b = True custom_op.parameter_map["hcom_parallel"].b = True + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") self.estimator_config.graph_options.rewrite_options.remapping = RewriterConfig.OFF self.estimator_config.graph_options.rewrite_options.memory_optimization = RewriterConfig.OFF custom_op.parameter_map["graph_run_mode"].i = 0 -- Gitee From c865f00a3c4213b7996486ee1093554f52a855b4 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:36:06 +0000 Subject: [PATCH 05/13] update TensorFlow/contrib/cv/VGG19_ID0374_for_TensorFlow/vgg19/trainer.py. --- .../contrib/cv/VGG19_ID0374_for_TensorFlow/vgg19/trainer.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/contrib/cv/VGG19_ID0374_for_TensorFlow/vgg19/trainer.py b/TensorFlow/contrib/cv/VGG19_ID0374_for_TensorFlow/vgg19/trainer.py index 79895712d..e7f574341 100644 --- a/TensorFlow/contrib/cv/VGG19_ID0374_for_TensorFlow/vgg19/trainer.py +++ b/TensorFlow/contrib/cv/VGG19_ID0374_for_TensorFlow/vgg19/trainer.py @@ -41,6 +41,7 @@ class Trainer(object): run_config = NPURunConfig( auto_tune_mode=os.getenv("FLAG_AUTOTUNE"), + auto_tune_mode="RL,GA", hcom_parallel=True, precision_mode="allow_mix_precision", enable_data_pre_proc=True, -- Gitee From e9049d0c993b8454a128e380f8782ac14e32ec3d Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:40:06 +0000 Subject: [PATCH 06/13] update TensorFlow/built-in/cv/image_classification/VGG19_for_TensorFlow/infer_from_pb.py. --- .../image_classification/VGG19_for_TensorFlow/infer_from_pb.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/cv/image_classification/VGG19_for_TensorFlow/infer_from_pb.py b/TensorFlow/built-in/cv/image_classification/VGG19_for_TensorFlow/infer_from_pb.py index ff04e440b..c1b2c82d2 100644 --- a/TensorFlow/built-in/cv/image_classification/VGG19_for_TensorFlow/infer_from_pb.py +++ b/TensorFlow/built-in/cv/image_classification/VGG19_for_TensorFlow/infer_from_pb.py @@ -79,6 +79,7 @@ class Classifier(object): # 4）set graph_run_mode=0，obtain better performance custom_op.parameter_map["graph_run_mode"].i = 0 + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") # -------------------------------------------------------------------------------- # load model， set graph input nodes and output nodes -- Gitee From 0b693c93829e787123669725921ee0bb4b81f342 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:43:06 +0000 Subject: [PATCH 07/13] cv/MobileFaceNet_Tensorflow/mobilefacenet-V2_ID0929_for_TensorFlow/train_nets.py. --- .../mobilefacenet-V2_ID0929_for_TensorFlow/train_nets.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/contrib/cv/MobileFaceNet_Tensorflow/mobilefacenet-V2_ID0929_for_TensorFlow/train_nets.py b/TensorFlow/contrib/cv/MobileFaceNet_Tensorflow/mobilefacenet-V2_ID0929_for_TensorFlow/train_nets.py index 87c79246f..c82e1cea0 100644 --- a/TensorFlow/contrib/cv/MobileFaceNet_Tensorflow/mobilefacenet-V2_ID0929_for_TensorFlow/train_nets.py +++ b/TensorFlow/contrib/cv/MobileFaceNet_Tensorflow/mobilefacenet-V2_ID0929_for_TensorFlow/train_nets.py @@ -210,6 +210,7 @@ if __name__ == '__main__': custom_op.parameter_map["fusion_switch_file"].s = \ tf.compat.as_bytes("./fusion_switch.cfg") custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes("allow_mix_precision") + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") config.graph_options.rewrite_options.remapping = RewriterConfig.OFF sess = tf.Session(config=config) sess.run(init) -- Gitee From 3d91d60e24dc17fdbefea043d51ad5364df5f933 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:45:37 +0000 Subject: [PATCH 08/13] built-in/cv/image_classification/InceptionV3_ID0491_for_TensorFlow/Incetpion_V3.py. --- .../InceptionV3_ID0491_for_TensorFlow/Incetpion_V3.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/TensorFlow/built-in/cv/image_classification/InceptionV3_ID0491_for_TensorFlow/Incetpion_V3.py b/TensorFlow/built-in/cv/image_classification/InceptionV3_ID0491_for_TensorFlow/Incetpion_V3.py index beda646d4..fc478bd1a 100644 --- a/TensorFlow/built-in/cv/image_classification/InceptionV3_ID0491_for_TensorFlow/Incetpion_V3.py +++ b/TensorFlow/built-in/cv/image_classification/InceptionV3_ID0491_for_TensorFlow/Incetpion_V3.py @@ -107,7 +107,7 @@ def run(): custom_op.name = "NpuOptimizer" custom_op.parameter_map["use_off_line"].b = True custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes("allow_mix_precision") - + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") config.graph_options.optimizer_options.global_jit_level = config_pb2.OptimizerOptions.OFF config.graph_options.rewrite_options.remapping = RewriterConfig.OFF if FLAGS.precision_mode == "allow_mix_precision": -- Gitee From da4585e10dff20be1a6de6997d3cc0ee45c1ef25 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:48:38 +0000 Subject: [PATCH 09/13] update TensorFlow/built-in/cv/detection/SSD-VGG_ID1619_for_TensorFlow/train.py. --- .../built-in/cv/detection/SSD-VGG_ID1619_for_TensorFlow/train.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/cv/detection/SSD-VGG_ID1619_for_TensorFlow/train.py b/TensorFlow/built-in/cv/detection/SSD-VGG_ID1619_for_TensorFlow/train.py index 10fd58af0..5b13e6ffb 100644 --- a/TensorFlow/built-in/cv/detection/SSD-VGG_ID1619_for_TensorFlow/train.py +++ b/TensorFlow/built-in/cv/detection/SSD-VGG_ID1619_for_TensorFlow/train.py @@ -208,6 +208,7 @@ def main(): custom_op.parameter_map["dynamic_graph_execute_mode"].s = tf.compat.as_bytes("lazy_recompile") custom_op.parameter_map["use_off_line"].b = True custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes(args.precision_mode) + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") if args.over_dump: custom_op.parameter_map["enable_dump_debug"].b = True custom_op.parameter_map["dump_debug_mode"].s = tf.compat.as_bytes("all") -- Gitee From 6cf8120eadfa48e2886bd5b17142689beea43e7c Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:52:32 +0000 Subject: [PATCH 10/13] update TensorFlow/built-in/cv/image_classification/MobileNetV1_ID0093_for_TensorFlow/train.py. --- .../MobileNetV1_ID0093_for_TensorFlow/train.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/cv/image_classification/MobileNetV1_ID0093_for_TensorFlow/train.py b/TensorFlow/built-in/cv/image_classification/MobileNetV1_ID0093_for_TensorFlow/train.py index acaf58d28..3f5394adb 100644 --- a/TensorFlow/built-in/cv/image_classification/MobileNetV1_ID0093_for_TensorFlow/train.py +++ b/TensorFlow/built-in/cv/image_classification/MobileNetV1_ID0093_for_TensorFlow/train.py @@ -127,6 +127,7 @@ def main(args): custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes("allow_mix_precision") custom_op.parameter_map["use_off_line"].b = True custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes(args.precision_mode) + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") config_proto.graph_options.rewrite_options.remapping = RewriterConfig.OFF sess = tf.Session(config=npu_config_proto(config_proto=config_proto)) sess.run(tf.global_variables_initializer()) -- Gitee From e1ca8597367ed5de12874a6a79030361eda326fc Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:55:36 +0000 Subject: [PATCH 11/13] built-in/cv/image_classification/MobileNetV3_ID0256_for_TensorFlow/train_cls-npu.py. --- .../MobileNetV3_ID0256_for_TensorFlow/train_cls-npu.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/cv/image_classification/MobileNetV3_ID0256_for_TensorFlow/train_cls-npu.py b/TensorFlow/built-in/cv/image_classification/MobileNetV3_ID0256_for_TensorFlow/train_cls-npu.py index dbb7fec0e..e051a8f8c 100644 --- a/TensorFlow/built-in/cv/image_classification/MobileNetV3_ID0256_for_TensorFlow/train_cls-npu.py +++ b/TensorFlow/built-in/cv/image_classification/MobileNetV3_ID0256_for_TensorFlow/train_cls-npu.py @@ -153,6 +153,7 @@ if __name__ == '__main__': custom_op = sess_config.graph_options.rewrite_options.custom_optimizers.add() custom_op.name = "NpuOptimizer" custom_op.parameter_map["use_off_line"].b = True + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") sess_config.graph_options.rewrite_options.remapping = RewriterConfig.OFF sess = tf.Session(config=sess_config) K.set_session(sess) -- Gitee From 53c9bba58fc84ea7f918a541e10b438382a097b8 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 03:57:38 +0000 Subject: [PATCH 12/13] update TensorFlow/built-in/cv/image_classification/AlexNet_ID0259_for_TensorFlow/train.py. --- .../image_classification/AlexNet_ID0259_for_TensorFlow/train.py | 1 + 1 file changed, 1 insertion(+) diff --git a/TensorFlow/built-in/cv/image_classification/AlexNet_ID0259_for_TensorFlow/train.py b/TensorFlow/built-in/cv/image_classification/AlexNet_ID0259_for_TensorFlow/train.py index 0c5296de4..ed0db8696 100644 --- a/TensorFlow/built-in/cv/image_classification/AlexNet_ID0259_for_TensorFlow/train.py +++ b/TensorFlow/built-in/cv/image_classification/AlexNet_ID0259_for_TensorFlow/train.py @@ -135,6 +135,7 @@ class AlexNet: custom_op.parameter_map["use_off_line"].b = True # ������ʾ�������ڕN��AI������ִ��ѵ�� custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes("allow_mix_precision") custom_op.parameter_map["hcom_parallel"].b = True + custom_op.parameter_map["auto_tune_mode"].s = tf.compat.as_bytes("RL,GA") config.graph_options.rewrite_options.remapping = RewriterConfig.OFF # ������ʾ�ر�remap bcast_op = broadcast_global_variables(0,1) sess = tf.Session(config=config) -- Gitee From f3e27f66ea4a737772f1b4f674a69f70cb63ad41 Mon Sep 17 00:00:00 2001 From: huangju1993 Date: Tue, 29 Mar 2022 04:08:39 +0000 Subject: [PATCH 13/13] built-in/cv/image_classification/EfficientNet-B4_ID0162_for_TensorFlow/main_npu.py. --- .../EfficientNet-B4_ID0162_for_TensorFlow/main_npu.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/TensorFlow/built-in/cv/image_classification/EfficientNet-B4_ID0162_for_TensorFlow/main_npu.py b/TensorFlow/built-in/cv/image_classification/EfficientNet-B4_ID0162_for_TensorFlow/main_npu.py index df5bad95c..3f97d3f13 100644 --- a/TensorFlow/built-in/cv/image_classification/EfficientNet-B4_ID0162_for_TensorFlow/main_npu.py +++ b/TensorFlow/built-in/cv/image_classification/EfficientNet-B4_ID0162_for_TensorFlow/main_npu.py @@ -714,7 +714,7 @@ def main(unused_argv): os.environ['TF_ENABLE_WINOGRAD_NONFUSED'] = '1' config = NPURunConfig( - #precision_mode="allow_fp32_to_fp16", + precision_mode="allow_mix_precision", #enable_data_pre_proc=True, save_checkpoints_steps=FLAGS.num_train_images // (FLAGS.train_batch_size * int(os.getenv('RANK_SIZE'))), session_config=estimator_config, -- Gitee