Merge pull request #2104 from HydrogenSulfate/add_ampO2_train

Add amp O2 train

ppcls/arch/backbone/model_zoo/vision_transformer.py

@@ -62,7 +62,7 @@ def drop_path(x, drop_prob=0., training=False):
         return x
     keep_prob = paddle.to_tensor(1 - drop_prob)
     shape = (paddle.shape(x)[0], ) + (1, ) * (x.ndim - 1)
-    random_tensor = keep_prob + paddle.rand(shape, dtype=x.dtype)
+    random_tensor = keep_prob + paddle.rand(shape).astype(x.dtype)
     random_tensor = paddle.floor(random_tensor)  # binarize
     output = x.divide(keep_prob) * random_tensor
     return output

ppcls/engine/evaluation/retrieval.py

@@ -159,6 +159,14 @@ def cal_feature(engine, name='gallery'):
         if len(batch) == 3:
             has_unique_id = True
             batch[2] = batch[2].reshape([-1, 1]).astype("int64")
+        if engine.amp and engine.amp_eval:
+            with paddle.amp.auto_cast(
+                    custom_black_list={
+                        "flatten_contiguous_range", "greater_than"
+                    },
+                    level=engine.amp_level):
+                out = engine.model(batch[0], batch[1])
+        else:
             out = engine.model(batch[0], batch[1])
         if "Student" in out:
             out = out["Student"]

ppcls/optimizer/optimizer.py

@@ -16,9 +16,9 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from paddle import optimizer as optim
-import paddle
+import inspect
 
+from paddle import optimizer as optim
 from ppcls.utils import logger
 
 
@@ -49,16 +49,27 @@ class SGD(object):
                  learning_rate=0.001,
                  weight_decay=None,
                  grad_clip=None,
+                 multi_precision=False,
                  name=None):
         self.learning_rate = learning_rate
         self.weight_decay = weight_decay
         self.grad_clip = grad_clip
+        self.multi_precision = multi_precision
         self.name = name
 
     def __call__(self, model_list):
         # model_list is None in static graph
         parameters = sum([m.parameters() for m in model_list],
                          []) if model_list else None
+        argspec = inspect.getargspec(optim.SGD.__init__).args
+        if 'multi_precision' in argspec:
+            opt = optim.SGD(learning_rate=self.learning_rate,
+                            parameters=parameters,
+                            weight_decay=self.weight_decay,
+                            grad_clip=self.grad_clip,
+                            multi_precision=self.multi_precision,
+                            name=self.name)
+        else:
             opt = optim.SGD(learning_rate=self.learning_rate,
                             parameters=parameters,
                             weight_decay=self.weight_decay,
@@ -242,8 +253,9 @@ class AdamW(object):
 
         if self.one_dim_param_no_weight_decay:
             self.no_weight_decay_param_name_list += [
-                p.name for model in model_list
-                for n, p in model.named_parameters() if len(p.shape) == 1
+                p.name
+                for model in model_list for n, p in model.named_parameters()
+                if len(p.shape) == 1
             ] if model_list else []
 
         opt = optim.AdamW(

test_tipc/config/GeneralRecognition/GeneralRecognition_PPLCNet_x2_5_train_amp_infer_python.txt → test_tipc/config/GeneralRecognition/GeneralRecognition_PPLCNet_x2_5_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/GeneralRecognition/GeneralRecognition_PPLCNet_x2_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/GeneralRecognition/GeneralRecognition_PPLCNet_x2_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/GeneralRecognition/GeneralRecognition_PPLCNet_x2_5.yaml
+eval:tools/eval.py -c ppcls/configs/GeneralRecognition/GeneralRecognition_PPLCNet_x2_5.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,14 +39,14 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_rec.py -c configs/inference_rec.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.rec_inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/Aliproduct/demo_test/
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null
 null:null

test_tipc/config/MobileNetV3/MobileNetV3_large_x1_0_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -3,7 +3,7 @@ model_name:MobileNetV3_large_x1_0
 python:python3.7
 gpu_list:0|0,1
 -o Global.device:gpu
--o Global.auto_cast:amp
+-o Global.auto_cast:null
 -o Global.epochs:lite_train_lite_infer=2|whole_train_whole_infer=120
 -o Global.output_dir:./output/
 -o DataLoader.Train.sampler.batch_size:8
@@ -12,16 +12,16 @@ train_model_name:latest
 train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
-trainer:norm_train|pact_train|fpgm_train
-norm_train:tools/train.py -c ppcls/configs/ImageNet/MobileNetV3/MobileNetV3_large_x1_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
-pact_train:tools/train.py -c ppcls/configs/slim/MobileNetV3_large_x1_0_quantization.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
-fpgm_train:tools/train.py -c ppcls/configs/slim/MobileNetV3_large_x1_0_prune.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
+trainer:amp_train
+amp_train:tools/train.py -c ppcls/configs/ImageNet/MobileNetV3/MobileNetV3_large_x1_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
+pact_train:null
+fpgm_train:null
 distill_train:null
 null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/MobileNetV3/MobileNetV3_large_x1_0.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/MobileNetV3/MobileNetV3_large_x1_0.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -33,20 +33,20 @@ fpgm_export:tools/export_model.py -c ppcls/configs/slim/MobileNetV3_large_x1_0_p
 distill_export:null
 kl_quant:deploy/slim/quant_post_static.py -c ppcls/configs/ImageNet/MobileNetV3/MobileNetV3_large_x1_0.yaml -o Global.save_inference_dir=./inference
 export2:null
-inference_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/whole_chain/MobileNetV3_large_x1_0_inference.tar
+pretrained_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/MobileNetV3_large_x1_0_pretrained.pdparams
 infer_model:../inference/
-infer_export:null
+infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null
 null:null

test_tipc/config/PPHGNet/PPHGNet_small_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:PPHGNet_small
+python:python3.7
+gpu_list:0|0,1
+-o Global.device:gpu
+-o Global.auto_cast:null
+-o Global.epochs:lite_train_lite_infer=2|whole_train_whole_infer=120
+-o Global.output_dir:./output/
+-o DataLoader.Train.sampler.batch_size:8
+-o Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./dataset/ILSVRC2012/val
+null:null
+##
+trainer:amp_train
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPHGNet/PPHGNet_small.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPHGNet/PPHGNet_small.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+null:null
+##
+===========================infer_params==========================
+-o Global.save_inference_dir:./inference
+-o Global.pretrained_model:
+norm_export:tools/export_model.py -c ppcls/configs/ImageNet/PPHGNet/PPHGNet_small.yaml
+quant_export:null
+fpgm_export:null
+distill_export:null
+kl_quant:null
+export2:null
+pretrained_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNet_small_pretrained.pdparams
+infer_model:../inference/
+infer_export:True
+infer_quant:Fasle
+inference:python/predict_cls.py -c configs/inference_cls.yaml
+-o Global.use_gpu:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
+-o Global.inference_model_dir:../inference
+-o Global.infer_imgs:../dataset/ILSVRC2012/val
+-o Global.save_log_path:null
+-o Global.benchmark:False
+null:null

test_tipc/config/PPHGNet/PPHGNet_tiny_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:PPHGNet_tiny
+python:python3.7
+gpu_list:0|0,1
+-o Global.device:gpu
+-o Global.auto_cast:null
+-o Global.epochs:lite_train_lite_infer=2|whole_train_whole_infer=120
+-o Global.output_dir:./output/
+-o DataLoader.Train.sampler.batch_size:8
+-o Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./dataset/ILSVRC2012/val
+null:null
+##
+trainer:amp_train
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPHGNet/PPHGNet_tiny.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPHGNet/PPHGNet_tiny.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+null:null
+##
+===========================infer_params==========================
+-o Global.save_inference_dir:./inference
+-o Global.pretrained_model:
+norm_export:tools/export_model.py -c ppcls/configs/ImageNet/PPHGNet/PPHGNet_tiny.yaml
+quant_export:null
+fpgm_export:null
+distill_export:null
+kl_quant:null
+export2:null
+pretrained_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNet_tiny_pretrained.pdparams
+infer_model:../inference/
+infer_export:True
+infer_quant:Fasle
+inference:python/predict_cls.py -c configs/inference_cls.yaml
+-o Global.use_gpu:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
+-o Global.inference_model_dir:../inference
+-o Global.infer_imgs:../dataset/ILSVRC2012/val
+-o Global.save_log_path:null
+-o Global.benchmark:False
+null:null

test_tipc/config/PPLCNet/PPLCNet_x0_25_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x0_25_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_25.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_25.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_25.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_25.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x0_35_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x0_35_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_35.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_35.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_35.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_35.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x0_5_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x0_5_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_5.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_5.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x0_75_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x0_75_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_75.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_75.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_75.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x0_75.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x1_0_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x1_0_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x1_5_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x1_5_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_5.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_5.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x2_0_train_amp_infer_python.txt → test_tipc/config/PPLCNet/PPLCNet_x2_0_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_0.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_0.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNet/PPLCNet_x2_5_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -3,7 +3,7 @@ model_name:PPLCNet_x2_5
 python:python3.7
 gpu_list:0|0,1
 -o Global.device:gpu
--o Global.auto_cast:amp
+-o Global.auto_cast:null
 -o Global.epochs:lite_train_lite_infer=2|whole_train_whole_infer=120
 -o Global.output_dir:./output/
 -o DataLoader.Train.sampler.batch_size:8
@@ -12,8 +12,8 @@ train_model_name:latest
 train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
-trainer:norm_train
-norm_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
+trainer:amp_train
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_5.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_5.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x2_5.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,13 +39,13 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null

test_tipc/config/PPLCNetV2/PPLCNetV2_base_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:PPLCNetV2_base
+python:python3.7
+gpu_list:0|0,1
+-o Global.device:gpu
+-o Global.auto_cast:null
+-o Global.epochs:lite_train_lite_infer=2|whole_train_whole_infer=120
+-o Global.output_dir:./output/
+-o DataLoader.Train.sampler.batch_size:null
+-o Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./dataset/ILSVRC2012/val
+null:null
+##
+trainer:amp_train
+amp_train:tools/train.py -c ppcls/configs/ImageNet/PPLCNetV2/PPLCNetV2_base.yaml -o Global.seed=1234 -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c ppcls/configs/ImageNet/PPLCNetV2/PPLCNetV2_base.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+null:null
+##
+===========================infer_params==========================
+-o Global.save_inference_dir:./inference
+-o Global.pretrained_model:
+norm_export:tools/export_model.py -c ppcls/configs/ImageNet/PPLCNetV2/PPLCNetV2_base.yaml
+quant_export:null
+fpgm_export:null
+distill_export:null
+kl_quant:null
+export2:null
+pretrained_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNetV2_base_pretrained.pdparams
+infer_model:../inference/
+infer_export:True
+infer_quant:Fasle
+inference:python/predict_cls.py -c configs/inference_cls.yaml
+-o Global.use_gpu:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
+-o Global.inference_model_dir:../inference
+-o Global.infer_imgs:../dataset/ILSVRC2012/val
+-o Global.save_log_path:null
+-o Global.benchmark:False
+null:null

test_tipc/config/ResNet/ResNet50_train_purefp16_infer_python.txt → test_tipc/config/ResNet/ResNet50_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/ResNet/ResNet50_amp_O1.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/ResNet/ResNet50.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ to_static_train:-o Global.to_static=True
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/ResNet/ResNet50.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/ResNet/ResNet50.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,15 +39,15 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null
 null:null
 ===========================train_benchmark_params==========================

test_tipc/config/ResNet/ResNet50_vd_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -3,7 +3,7 @@ model_name:ResNet50_vd
 python:python3.7
 gpu_list:0|0,1
 -o Global.device:gpu
--o Global.auto_cast:amp
+-o Global.auto_cast:null
 -o Global.epochs:lite_train_lite_infer=2|whole_train_whole_infer=120
 -o Global.output_dir:./output/
 -o DataLoader.Train.sampler.batch_size:8
@@ -12,16 +12,16 @@ train_model_name:latest
 train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
-trainer:norm_train|pact_train|fpgm_train
-norm_train:tools/train.py -c ppcls/configs/ImageNet/ResNet/ResNet50_vd.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
-pact_train:tools/train.py -c ppcls/configs/slim/ResNet50_vd_quantization.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
-fpgm_train:tools/train.py -c ppcls/configs/slim/ResNet50_vd_prune.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False
+trainer:amp_train
+amp_train:tools/train.py -c ppcls/configs/ImageNet/ResNet/ResNet50_vd.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
+pact_train:null
+fpgm_train:null
 distill_train:null
 null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/ResNet/ResNet50_vd.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/ResNet/ResNet50_vd.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -33,20 +33,20 @@ fpgm_export:tools/export_model.py -c ppcls/configs/slim/ResNet50_vd_prune.yaml
 distill_export:null
 kl_quant:deploy/slim/quant_post_static.py -c ppcls/configs/ImageNet/ResNet/ResNet50_vd.yaml -o Global.save_inference_dir=./inference
 export2:null
-inference_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/whole_chain/ResNet50_vd_inference.tar
+pretrained_model_url:https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/ResNet50_vd_pretrained.pdparams
 infer_model:../inference/
-infer_export:null
+infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null
 null:null

test_tipc/config/SwinTransformer/SwinTransformer_tiny_patch4_window7_224_train_amp_infer_python.txt → test_tipc/config/SwinTransformer/SwinTransformer_tiny_patch4_window7_224_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./dataset/ILSVRC2012/val
 null:null
 ##
 trainer:amp_train
-amp_train:tools/train.py -c ppcls/configs/ImageNet/SwinTransformer/SwinTransformer_tiny_patch4_window7_224.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=128 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
+amp_train:tools/train.py -c ppcls/configs/ImageNet/SwinTransformer/SwinTransformer_tiny_patch4_window7_224.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c ppcls/configs/ImageNet/SwinTransformer/SwinTransformer_tiny_patch4_window7_224.yaml
+eval:tools/eval.py -c ppcls/configs/ImageNet/SwinTransformer/SwinTransformer_tiny_patch4_window7_224.yaml -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2
 null:null
 ##
 ===========================infer_params==========================
@@ -39,14 +39,14 @@ infer_export:True
 infer_quant:Fasle
 inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.use_gpu:True|False
--o Global.enable_mkldnn:True|False
--o Global.cpu_num_threads:1|6
--o Global.batch_size:1|16
--o Global.use_tensorrt:True|False
--o Global.use_fp16:True|False
+-o Global.enable_mkldnn:False
+-o Global.cpu_num_threads:6
+-o Global.batch_size:1
+-o Global.use_tensorrt:False
+-o Global.use_fp16:False
 -o Global.inference_model_dir:../inference
 -o Global.infer_imgs:../dataset/ILSVRC2012/val
 -o Global.save_log_path:null
--o Global.benchmark:True
+-o Global.benchmark:False
 null:null
 null:null