Skip to content

Add flags to control whether to check Nan value of hccl_allreduce_sum… #35298

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 31, 2021
Merged

Add flags to control whether to check Nan value of hccl_allreduce_sum… #35298

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 10 additions & 3 deletions paddle/fluid/operators/collective/c_allreduce_op.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,10 @@ limitations under the License. */
#include "paddle/fluid/platform/hccl_helper.h"
#endif

#if defined(PADDLE_WITH_ASCEND_CL)
DECLARE_bool(hccl_check_nan);
#endif

namespace paddle {
namespace operators {

Expand Down Expand Up @@ -140,6 +144,7 @@ inline bool ContainsNan(const paddle::platform::NPUDeviceContext& dev_ctx,
try {
const auto& runner_mean = paddle::operators::NpuOpRunner(
"ReduceMeanD", {*in}, {mean}, {{"axes", axes}, {"keep_dims", false}});
runner_mean.Run(stream);
TensorToVector(mean, dev_ctx, &vec);
} catch (...) {
LOG(WARNING) << "ContainsNan catch exception";
Expand Down Expand Up @@ -233,9 +238,11 @@ class CAllReduceOpASCENDKernel : public framework::OpKernel<T> {
break;
}
case framework::proto::VarType::FP32: {
VLOG(4) << "prepare to FoundNanInf";
found_nan = ContainsNan(*dev_ctx, dev_ctx->stream(), in);
VLOG(4) << "check_numerics:" << found_nan;
if (FLAGS_hccl_check_nan) {
VLOG(3) << "prepare to FoundNanInf";
found_nan = ContainsNan(*dev_ctx, dev_ctx->stream(), in);
VLOG(3) << "check_numerics:" << found_nan;
}
break;
}
default:
Expand Down
3 changes: 3 additions & 0 deletions paddle/fluid/platform/flags.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,6 +93,9 @@ DEFINE_string(selected_npus, "",
"This option is useful when doing multi process training and "
"each process have only one device (NPU). If you want to use "
"all visible devices, set this to empty string.");
DEFINE_bool(hccl_check_nan, false,
"Check Nan in tensor before hccl_allreduce_sum otherwise it'll "
"core when meets Nan value");
DEFINE_string(
npu_config_path, "",
"The absolute path of configuration json file, like: /tmp/config.json. "
Expand Down
1 change: 1 addition & 0 deletions python/paddle/fluid/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -248,6 +248,7 @@ def __bootstrap__():
'gpu_memory_limit_mb',
'npu_config_path',
'get_host_by_name_time',
'hccl_check_nan',
]

core.init_gflags(["--tryfromenv=" + ",".join(read_env_flags)])
Expand Down