feat(ppsci): support data_effient_nopt for training and test

Author: xiaoyewww

examples/data_efficient_nopt/config/operators_poisson.yaml

@@ -31,7 +31,7 @@ default: &DEFAULT
   entity: 'entity_name'
   project: 'proj_name'
   group: 'poisson'
-  log_to_wandb: !!bool True
+  log_to_wandb: !!bool False
   distill: !!bool False
   subsample: 1
   exp_dir: './exp/'
@@ -84,7 +84,7 @@ poisson-64-scale-e5_15: &poisson_64_e5_15
   scales_path: '/home/aistudio/data_efficient_nopt/data/possion_64/poisson_64_e5_15_train_scale.npy'
   train_rand_idx_path: '/home/aistudio/data_efficient_nopt/data/possion_64/train_rand_idx.npy'
   batch_size: 128
-  log_to_wandb: !!bool True
+  log_to_wandb: !!bool False
   learning_rate: 1E-3
 
   mode_cut: 32
@@ -107,7 +107,7 @@ pois-64-pretrain-e1_20: &pois_64_e1_20_pt
   scales_path:  '/home/aistudio/data_efficient_nopt/data/possion_64/poisson_64_e1_20_train_scale.npy'
   train_rand_idx_path: '/home/aistudio/data_efficient_nopt/data/possion_64/train_rand_idx.npy'
   batch_size: 128
-  log_to_wandb: !!bool True
+  log_to_wandb: !!bool False
   mode_cut: 32
   embed_cut: 64
   fc_cut: 2
@@ -128,7 +128,7 @@ pois-64-finetune-e5_15: &pois_64_e5_15_ft
   scales_path:  '/home/aistudio/data_efficient_nopt/data/possion_64/poisson_64_e5_15_train_scale.npy'
   train_rand_idx_path: '/home/aistudio/data_efficient_nopt/data/possion_64/train_rand_idx.npy'
   batch_size: 128
-  log_to_wandb: !!bool True
+  log_to_wandb: !!bool False
   mode_cut: 32
   embed_cut: 64
   fc_cut: 2

examples/data_efficient_nopt/config/vmae_config_pretrain.yaml

@@ -1,6 +1,6 @@
 basic_config: &basic_config
   # Run settings
-  log_to_wandb: !!bool True # Use wandb integration
+  log_to_wandb: !!bool False # Use wandb integration
   log_to_screen: !!bool True # Log progress to screen.
   save_checkpoint: !!bool True # Save checkpoints
   checkpoint_save_interval: 100 # Save every # epochs - also saves "best" according to val loss

examples/data_efficient_nopt/pretrain_basic.py

@@ -91,7 +91,7 @@ def add_weight_decay(model, weight_decay=1e-5, inner_lr=1e-3, skip_list=()):
     decay = []
     no_decay = []
     for name, param in model.named_parameters():
-        if not param.requires_grad:
+        if param.stop_gradient:
             continue
         if len(param.squeeze().shape) <= 1 or name in skip_list:
             no_decay.append(param)
@@ -230,24 +230,32 @@ def initialize_optimizer(self, params):
         if params.optimizer == "adam":
             if self.params.learning_rate < 0:
                 self.optimizer = DAdaptAdam(
-                    parameters, lr=1.0, growth_rate=1.05, log_every=100, decouple=True
+                    parameters,
+                    learning_rate=1.0,
+                    growth_rate=1.05,
+                    log_every=100,
+                    decouple=True,
                 )
             else:
-                self.optimizer = optim.AdamW(parameters, lr=params.learning_rate)
+                self.optimizer = optim.AdamW(
+                    parameters=parameters, learning_rate=params.learning_rate
+                )
         elif params.optimizer == "adan":
             # if self.params.learning_rate < 0:
-            #     self.optimizer =  DAdaptAdan(parameters, lr=1., growth_rate=1.05, log_every=100)
+            #     self.optimizer =  DAdaptAdan(parameters, learning_rate=1., growth_rate=1.05, log_every=100)
             # else:
-            #     self.optimizer = Adan(parameters, lr=params.learning_rate)
+            #     self.optimizer = Adan(parameters, learning_rate=params.learning_rate)
             raise NotImplementedError("Adan not implemented yet")
         elif params.optimizer == "sgd":
             self.optimizer = optim.SGD(
-                self.model.parameters(), lr=params.learning_rate, momentum=0.9
+                parameters=self.model.parameters(),
+                learning_rate=params.learning_rate,
+                momentum=0.9,
             )
         else:
             raise ValueError(f"Optimizer {params.optimizer} not supported")
         self.gscaler = amp.GradScaler(
-            enabled=(self.mp_type == paddle.float16 and params.enable_amp)
+            enable=(self.mp_type == paddle.float16 and params.enable_amp)
         )
 
     def initialize_scheduler(self, params):