search

larq / compute-engine

Highly optimized inference engine for Binarized Neural Networks
https://docs.larq.dev/compute-engine
Apache License 2.0
242 stars 34 forks source link

:arrow_up: TensorFlow 2.9 #724

Closed lgeiger closed 2 years ago

lgeiger commented 2 years ago

What do these changes do?

This PR updates TensorFlow to version 2.9.0.

How Has This Been Tested?

CI and triggered a test release.

Benchmark Results

TODO

Tombana commented 2 years ago

I compared the benchmark binaries produced by CI (through bazel, not cmake) from the TF2.8 release with the TF2.9 test-release from this PR. I used a Raspberry Pi 4B at 1.5 GHz with QuickNet from larq-zoo in float. It uses XNNPack for the non-binary layers. For single-threaded benchmarks I found 46.4 ms for both builds, and for 4-threaded benchmarks I found 21.7 ms in both cases. I conclude that the benchmark results did not change with this TF 2.9 upgrade.