Chanjun
i don't want to re-install tensorflow when some tf header changed ( in my case , may changed by myself )？
rt
https://github.com/DeepRec-AI/DeepRec/blob/8d8e16aae66add22cf8a4812d549c83f3569ef13/addons/sparse_operation_kit/core/adapter/lookup_adapter.cpp#L71
run_sok_horovod.py use GradientTape, but tf1.15 has no GradientTape func? can provide 1.15 example with sok?
how to fix it?
Like ipairs(t), but stops at table.maxn(t) instead of the first nil value.
can use one big kernel rather than many small kernel? maybe one kernel faster?