diff --git a/paddle/fluid/framework/parallel_executor.cc b/paddle/fluid/framework/parallel_executor.cc index 1a2e6a5f8676d493c5476fd1d7b6b4bcf1c33aed..b78dc3b8ae2b33fa872bf971153d844428e797c1 100644 --- a/paddle/fluid/framework/parallel_executor.cc +++ b/paddle/fluid/framework/parallel_executor.cc @@ -131,6 +131,7 @@ struct ScaleLossGradOpHandle : public OpHandle { : coeff_(static_cast(1.0 / num_dev)), scope_(scope), place_(place) { + cudaSetDevice(boost::get(place_).device); PADDLE_ENFORCE(cudaEventCreateWithFlags(&ev_, cudaEventDisableTiming)); VLOG(3) << "Create " << ev_; }