Hi there,
I’m trying to compile DeepSpeech/specific Tensorflow on my aarch64 machine following this guide.
The problem I have is that the build script can’t find package for @com_github_nelhage_rules_boost//:boost/boost.bzl.
Here is the log.
What have I done wrong?
Thanks.
This was added after 0.7.4, so it’s likely you are not building that version.
Please understand we don’t support native Aarch64 build, only cross-compilation, so it’s likely you will have some things to fix.
What is your usecase that forces you to rebuild instead of using our prebuilt binaries?
We also only support building and running TFLite runtime on Aarch64.
This was added after 0.7.4, so it’s likely you are not building that version.
Yes, the link to build 0.8.x was broken so I had to use an older documentation.
What is your usecase that forces you to rebuild instead of using our prebuilt binaries?
Well, I tried to install DeepSpeech for aarch64, but pip3 install deepspeech did not find anything. --verbose told me later that there were no binaries available for aarch64 (only x86_64, amd, armv7 and so on).
We also only support building and running TFLite runtime on Aarch64.
There are now really powerful aarch64 machines, and some of them even sport a NPU.
Thanks.
Please pip install https://github.com/mozilla/STT/releases/download/v0.7.4/deepspeech-0.7.4-cp37-cp37m-linux_aarch64.whl we can’t upload aarch64 wheels on PyPi
It’s not about that, it’s about the tensorflow runtime itself, only tflite part is really optimized for Aarch64. Also, NPU might not be useful for our model: NNAPI and other often do not provide tangible speedup and require invasive changes to the model.
BTW we shipped 0.8.1 that includes faster TFLite: https://github.com/mozilla/STT/releases/download/v0.8.1/deepspeech-0.8.1-cp37-cp37m-linux_aarch64.whl and it can use previous models.
Thanks a lot, I will give it a try.