In this meetup David Gruzman from www.nestlogic.com will present how Tensorflow is built internally. We will cover latest version 2.0
This meetup is not aimed as "introduction to TF" but more to improve understanding of this beautiful engineering work.
In more details
- how work is separated between client and core
- define by run vs define and run paradigms
- what is API between client and core
- what are ways to extend TF
- what optimizations it have internally
- how python code became high performance distributed computational graph
- how high performance are achieved
- how Horovod is built, to show alternative way to distribution
- other things
you welcome to suggest other TF related topics to cover. Short in-depth presentation about some aspect of TF can be very welcome
18.00 - gathering and networking
18.30 - 20.00 presentation with short break.