update setup.py

fleetrec/core/factory.py

@@ -32,7 +32,10 @@ class TrainerFactory(object):
     def _build_trainer(config, yaml_path):
         print(envs.pretty_print_envs(envs.get_global_envs()))
 
-        train_mode = envs.get_runtime_envion("train.trainer")
+        train_mode = envs.get_global_env("train.strategy.mode")
+
+        if train_mode is not None:
+            train_mode = envs.get_runtime_envion("train.trainer")
 
         if train_mode == "SingleTraining":
             trainer = SingleTrainer(yaml_path)

fleetrec/core/trainers/cluster_trainer.py

@@ -44,6 +44,12 @@ class ClusterTrainer(TranspileTrainer):
 
     def build_strategy(self):
         mode = envs.get_global_env("train.strategy.mode")
+
+        if mode is None:
+            mode = envs.get_runtime_envion("train.strategy.mode")
+
+        assert mode is not None
+
         strategy = None
 
         if mode == "async":

fleetrec/examples/build_in/cluster_training_local.yaml

-
-trainer:  "LocalClusterTraining"
-
-pserver_num: 2
-trainer_num: 2
-start_port: 36001
-log_dirname: "logs"
-
-strategy:
-  mode: "async"

fleetrec/examples/build_in/ctr-dnn_train.yaml

@@ -15,7 +15,6 @@
 train:
   threads: 12
   epochs: 10
-  trainer:  "single_training.yaml"
 
   reader:
     mode: "dataset"

fleetrec/run.py

@@ -60,6 +60,7 @@ if __name__ == "__main__":
         cluster_envs["start_port"] = 36001
         cluster_envs["log_dir"] = "logs"
         cluster_envs["train.trainer"] = "ClusterTraining"
+        cluster_envs["train.strategy.mode"] = "async"
 
         local_cluster_engine(cluster_envs, args.model)
     elif args.engine == "LocalMPI":