stylable
stylable copied to clipboard
How does the tensorflow scheduler plugin used in the tf_benchmark_cnn.py
Hi, I'm testing the bytescheduler in tensorflow, follow the tutorial, compile the libplugin.so, and load it in the benchmark script. Now, I load the library in another script (my own), but i found that the training fps is almost same, so i wonder where does the scheduler algorithm use in code, it's explicit or explicit?Thanks.
If the plugin does not run, check https://github.com/bytedance/byteps/tree/bytescheduler/bytescheduler#tensorflow . If the plugin runs successfully but with no distributed training speedup, you can profile the trace and see if how much is potential improvement space and whether communication scheduling can help.