diff --git a/python/orca/src/bigdl/orca/learn/pytorch/estimator.py b/python/orca/src/bigdl/orca/learn/pytorch/estimator.py
index 98e2876c301..0e36e11c7d3 100644
--- a/python/orca/src/bigdl/orca/learn/pytorch/estimator.py
+++ b/python/orca/src/bigdl/orca/learn/pytorch/estimator.py
@@ -55,6 +55,7 @@ def from_torch(*,
                    config=None,
                    scheduler_step_freq="batch",
                    use_tqdm=False,
+                   workers_per_node=1,
                    backend="horovod"):
         if backend == "horovod":
             return PyTorchHorovodEstimatorWrapper(model_creator=model,
@@ -65,7 +66,8 @@ def from_torch(*,
                                                   initialization_hook=initialization_hook,
                                                   config=config,
                                                   scheduler_step_freq=scheduler_step_freq,
-                                                  use_tqdm=use_tqdm)
+                                                  use_tqdm=use_tqdm,
+                                                  workers_per_node=workers_per_node)
         elif backend == "bigdl":
             return PytorchSparkEstimatorWrapper(model=model,
                                                 loss=loss,
@@ -87,7 +89,8 @@ def __init__(self,
                  initialization_hook=None,
                  config=None,
                  scheduler_step_freq="batch",
-                 use_tqdm=False):
+                 use_tqdm=False,
+                 workers_per_node=1):
         from zoo.orca.learn.pytorch.pytorch_horovod_estimator import PyTorchHorovodEstimator
         self.estimator = PyTorchHorovodEstimator(model_creator=model_creator,
                                                  optimizer_creator=optimizer_creator,
@@ -97,7 +100,8 @@ def __init__(self,
                                                  initialization_hook=initialization_hook,
                                                  config=config,
                                                  scheduler_step_freq=scheduler_step_freq,
-                                                 use_tqdm=use_tqdm)
+                                                 use_tqdm=use_tqdm,
+                                                 workers_per_node=workers_per_node)
 
     def fit(self, data, epochs=1, profile=False, reduce_results=True, info=None):
         """