Installation frustrations

DeepSpeech
#1

Please note the following error I get when I’ve tried to use deepsearch from a pip install and also when running the prebuilt deepsearch downloaded using taskcluster.py:

tensorflow/core/platform/cpu_feature_guard.cc:35] The TensorFlow library was compiled to use AVX2 instructions, but these aren’t available on your machine.

I realize this is your first release but outside of mozillains, how are people managing to install deepsearch and make it run on a workstation? I have yet to find instructions containing clear steps to realizing a working deepsearch.

A section titled #Installation, in any README, ought to contain all of the required installation instructions. Neither the root README nor native_client/README have them.

#2

As I said on the other thread, there is so far no other alternative than rebuilding libtensorflow_cc.so and re-producing Python bindings in this case. I understand this is frustrating, but so far there is no runtime-level enabling/disabling of those optimizations, and we enabled them.

#3

And there is no “Installation section”, because that is documented right at the top of the root’s README.md: https://github.com/mozilla/DeepSpeech#project-deepspeech

You can see the pip install deepspeech statement, and as I said, README.md in native_client/ documents how to rebuild things in your case.

#4

I have the same problem and would love to see someone provide binaries for non-AVX2 (and non-NVidia) platforms. Not really in the mood for lengthy compilation sessions.

Cheers,
Stefan Reich
BotCompany.de

#5

Thanks for your feedback, I guess we’ll have to try and find a solution soon :slight_smile:

#6

Do you solve the problem?I get it , I can not solve it?thanks!!

#7

We are publishing AVX and non AVX2+FMA binaries now, from TaskCluster only for the moment: https://tools.taskcluster.net/index/artifacts/project.deepspeech.deepspeech.native_client.master/cpu