sxjscience commented on pull request #18987: URL: https://github.com/apache/incubator-mxnet/pull/18987#issuecomment-679454243
I think CNNs are generally static shape while models in NLP are generally dynamic shape. Do we have any plan for improving the memory usage? Get Outlook for iOS<https://aka.ms/o0ukef> ________________________________ From: Sheng Zha <[email protected]> Sent: Monday, August 24, 2020 6:30:41 PM To: apache/incubator-mxnet <[email protected]> Cc: Xingjian SHI <[email protected]>; Mention <[email protected]> Subject: Re: [apache/incubator-mxnet] [ENV] update runtime setting default values (#18987) @szha commented on this pull request. ________________________________ In src/storage/storage.cc<https://github.com/apache/incubator-mxnet/pull/18987#discussion_r476040201>: > @@ -67,8 +67,9 @@ StorageManager *CreateStorageManager(const Context &ctx, const char *context, int num_gpu_device, std::string *pStrategy) { const auto env_var = env_var_name(context, pool_type); const char *type = getenv(env_var.c_str()); - if (type == nullptr) - type = "Naive"; // default pool + if (type == nullptr) { + type = "Round"; // default pool My hope is of course to provide a good out-of-the-box usage experience to mxnet users. From what I observed, there seems to be more models with dynamic shape inputs than static ones, and many of the static-shape models can still run in this setting, hence the proposal. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<https://github.com/apache/incubator-mxnet/pull/18987#discussion_r476040201>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABHQH3TMASLXDEBXVJXNACLSCMH4DANCNFSM4QI2O3KQ>. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
