synesthesiam
commented
3 years ago Hi, @MarkinBrisbane. Thanks for giving it a try!
I've tried and (so far) failed to build PyTorch without AVX instructions. I thought I had succeeded originally because the -noavx versions ran much slower on my CPU (with AVX), but when I finally tried it on a system without AVX (Pentium G440) it still crashed.
I spent a week trying various build options, both for PyTorch's setup.py and for its underlying CMake build. No matter what I try, I can't seem to stop the compiler from inserting AVX instructions somewhere. I've even built it on the Pentium G440, and it still adds them!
If anyone has any ideas, I'd love to hear them :)
MarkinBrisbane
Hi @synesthesiam (Michael), I stumbled across your work whilst trying to learn more about AI/TTS so eagerly tried to run it in docker which lives in a guest on my mini-vm server which has a Celeron J4105 processor. I followed the directions to use the synesthesiam/mozillatts:en-noavx tag which results in the container loading and the webserver running (ports mapped 5002:5002).
Not long after I enter some text and select "Speak" it returns Error: Failed to fetch
The container then shows as stopped (I am lazy and use portainer)
Container status: Stopped for 25 minutes with exit code 132
End of log file: [INFO] * Running on http://0.0.0.0:5002/ (Press CTRL+C to quit) [INFO] 192.168.1.110 - - [17/Feb/2021 05:46:48] "GET / HTTP/1.1" 200 - [INFO] Synthesizing (4 char(s))... [INFO] initializing backend espeak-1.48.03 /run.sh: line 3: 6 Illegal instruction (core dumped) bin/python3 -m tts_web "$@"