testing

Author: typhoonzero

paddle/fluid/platform/nccl_helper.h

@@ -108,8 +108,8 @@ struct NCCLContextMap {
         for (auto &gpu_id : order_) {
           int rank = trainer_id * order_.size() + gpu_id;
           PADDLE_ENFORCE(cudaSetDevice(gpu_id));
-          PADDLE_ENFORCE(
-              ncclCommInitRank(comms.get() + gpu_id, nranks, *nccl_id, rank));
+          PADDLE_ENFORCE(platform::dynload::ncclCommInitRank(
+              comms.get() + gpu_id, nranks, *nccl_id, rank));
         }
       }
     }

python/paddle/fluid/parallel_executor.py

@@ -30,7 +30,9 @@ def __init__(self,
                  num_threads=None,
                  allow_op_delay=False,
                  share_vars_from=None,
-                 use_default_grad_scale=True):
+                 use_default_grad_scale=True,
+                 num_nodes=0,
+                 trainer_id=0):
         """
         ParallelExecutor can run program in parallel.
 
@@ -129,7 +131,9 @@ def __init__(self,
             scope,
             local_scopes,
             allow_op_delay,
-            use_default_grad_scale)
+            use_default_grad_scale,
+            num_nodes,
+            trainer_id)
         self.scope = scope
 
     def run(self, fetch_list, feed=None, feed_dict=None):