未验证 提交 8347354d 编写于 作者: R Ruibiao Chen 提交者: GitHub

Do not release memory cache after build_op_func_list in interpretercore (#47910)

上级 81e16a85
...@@ -813,12 +813,6 @@ void BuildOpFuncList(const platform::Place& place, ...@@ -813,12 +813,6 @@ void BuildOpFuncList(const platform::Place& place,
interpreter::LogDeviceMemoryStats(place); interpreter::LogDeviceMemoryStats(place);
} }
// NOTE(Ruibiao): Release memory cache to avoid memory fragments in Allocator.
// It reduce about 10% memory usage for V100 8-GPU training of
// transformer_base_bs4096_amp_fp16 and transformer_base_bs4096_pure_fp16
// model.
memory::Release(place);
} }
void LogDeviceMemoryStats(const platform::Place& place) { void LogDeviceMemoryStats(const platform::Place& place) {
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册