This message was deleted.

Hi @Tycho, parallelism in Tensorflow is already set by default with

TF_NUM_INTRAOP_THREADS

and

TF_NUM_INTEROP_THREADS

. The default parallelism is set to equal to the number of CPU cores. I think believe the code allows override at the moment. Could you please share why you need it configurable?

Hello Sean, for performance tests.

Got it. Currently the framework sets it automatically based on CPU cores. In our experience, setting them equal to the CPU cores typically yield the best results. Perhaps you can allocate different amount of CPU to your containerized image