I’ve built wheels for TensorFlow r1.12.0rc1 and the latest version of DeepSpeech (Python 3.6 only, sorry if you’re still using 2.7), and would be happy to share them if I had a place to upload to. Any suggestions?
The source for the TensorFlow fork is at https://github.com/rgobbel/tensorflow/tree/r1.12-agx_xavier. The changes from the default r1.12 branch are pretty minimal. Note that I’ve also tested this only minimally, though I haven’t run into any problems. The AWS SDK changes are only for Jetson and would probably cause builds on other platforms to break, since it simply uses the default case to add linux-shared/*.cpp to the artifact.
Building on the AGX Xavier takes several hours. The default instructions still apply. My environment on the Xavier is based on JetPack 4.1 DP.
set_config_vars.sh sets all variables requested by the configure script, but assumes a virtual environment based on Python 3.6, so you might need to modify it for your world. To use it:
. ./set_config_vars.sh ./configure