!8705 modify lr and epoch for 8p

From: @wukesong
Reviewed-by: @yingjy,@oacjiewen
Signed-off-by: @yingjy

model_zoo/official/cv/alexnet/src/config.py

@@ -51,12 +51,4 @@ alexnet_imagenet_cfg = edict({
 
     # lr
     'is_dynamic_loss_scale': 0,
-    'label_smooth': 1,
-    'label_smooth_factor': 0.1,
-    'lr_scheduler': 'cosine_annealing',
-    'warmup_epochs': 5,
-    'lr_epochs': [30, 60, 90, 120],
-    'lr_gamma': 0.1,
-    'T_max': 150,
-    'eta_min': 0.0,
 })

model_zoo/official/cv/alexnet/src/generator_lr.py

@@ -14,7 +14,6 @@
 # ============================================================================
 """learning rate generator"""
 import math
-from collections import Counter
 import numpy as np
 
 def get_lr_cifar10(current_step, lr_max, total_epochs, steps_per_epoch):
@@ -44,71 +43,18 @@ def get_lr_cifar10(current_step, lr_max, total_epochs, steps_per_epoch):
 
     return learning_rate
 
-def get_lr_imagenet(cfg, steps_per_epoch):
+def get_lr_imagenet(lr, epoch_size, steps_per_epoch):
     """generate learning rate array"""
-    if cfg.lr_scheduler == 'exponential':
-        lr = warmup_step_lr(cfg.learning_rate,
-                            cfg.lr_epochs,
-                            steps_per_epoch,
-                            cfg.warmup_epochs,
-                            cfg.epoch_size,
-                            gamma=cfg.lr_gamma,
-                            )
-    elif cfg.lr_scheduler == 'cosine_annealing':
-        lr = warmup_cosine_annealing_lr(cfg.learning_rate,
-                                        steps_per_epoch,
-                                        cfg.warmup_epochs,
-                                        cfg.epoch_size,
-                                        cfg.T_max,
-                                        cfg.eta_min)
-    else:
-        raise NotImplementedError(cfg.lr_scheduler)
-
+    lr = warmup_cosine_annealing_lr(lr, epoch_size, steps_per_epoch)
     return lr
 
-
-
 def linear_warmup_lr(current_step, warmup_steps, base_lr, init_lr):
     """Linear learning rate"""
     lr_inc = (float(base_lr) - float(init_lr)) / float(warmup_steps)
     lr = float(init_lr) + lr_inc * current_step
     return lr
 
-def warmup_step_lr(lr, lr_epochs, steps_per_epoch, warmup_epochs, max_epoch, gamma=0.1):
-    """Linear warm up learning rate"""
-    base_lr = lr
-    warmup_init_lr = 0
-    total_steps = int(max_epoch * steps_per_epoch)
-    warmup_steps = int(warmup_epochs * steps_per_epoch)
-    milestones = lr_epochs
-    milestones_steps = []
-    for milestone in milestones:
-        milestones_step = milestone * steps_per_epoch
-        milestones_steps.append(milestones_step)
-
-    lr_each_step = []
-    lr = base_lr
-    milestones_steps_counter = Counter(milestones_steps)
-    for i in range(total_steps):
-        if i < warmup_steps:
-            lr = linear_warmup_lr(i + 1, warmup_steps, base_lr, warmup_init_lr)
-        else:
-            lr = lr * gamma**milestones_steps_counter[i]
-        lr_each_step.append(lr)
-
-    return np.array(lr_each_step).astype(np.float32)
-
-def multi_step_lr(lr, milestones, steps_per_epoch, max_epoch, gamma=0.1):
-    return warmup_step_lr(lr, milestones, steps_per_epoch, 0, max_epoch, gamma=gamma)
-
-def step_lr(lr, epoch_size, steps_per_epoch, max_epoch, gamma=0.1):
-    lr_epochs = []
-    for i in range(1, max_epoch):
-        if i % epoch_size == 0:
-            lr_epochs.append(i)
-    return multi_step_lr(lr, lr_epochs, steps_per_epoch, max_epoch, gamma=gamma)
-
-def warmup_cosine_annealing_lr(lr, steps_per_epoch, warmup_epochs, max_epoch, T_max, eta_min=0):
+def warmup_cosine_annealing_lr(lr, max_epoch, steps_per_epoch, warmup_epochs=5, T_max=150, eta_min=0.0):
     """ Cosine annealing learning rate"""
     base_lr = lr
     warmup_init_lr = 0

model_zoo/official/cv/alexnet/train.py

@@ -71,6 +71,8 @@ if __name__ == "__main__":
         context.set_context(device_id=args.device_id)
 
         if device_num > 1:
+            cfg.learning_rate = cfg.learning_rate * device_num
+            cfg.epoch_size = cfg.epoch_size * 2
             context.reset_auto_parallel_context()
             context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
                                               gradients_mean=True)
@@ -107,7 +109,7 @@ if __name__ == "__main__":
 
     elif args.dataset_name == 'imagenet':
         loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
-        lr = Tensor(get_lr_imagenet(cfg, step_per_epoch))
+        lr = Tensor(get_lr_imagenet(cfg.learning_rate, cfg.epoch_size, step_per_epoch))
         opt = nn.Momentum(params=get_param_groups(network),
                           learning_rate=lr,
                           momentum=cfg.momentum,