0

I'm trying to run the preview version of a Computer Vision docker container on an Red Hat Enterprise Linux Server release 7.5 on-premises. I've pulled the docker container containerpreview.azurecr.io/microsoft/cognitive-services-read:latest and run like this

docker run --rm -it -p 5000:5000 --memory 16g --cpus 8
containerpreview.azurecr.io/microsoft/cognitive-services-read
Eula=accept Billing={ENDPOINT} ApiKey={API_KEY}

the service is up, the swagger is visible and the status endpoint returns OK.

However when I try to use the /vision/v2.0/read/core/Analyze endpoint the machine give me this log

Initialize on-prem Read 2.0 GA...
/var/tmp/scleXV71Y: line 8:    10 Illegal instruction     (core dumped) dotnet Microsoft.CloudAI.Containers.OneOcr.2.0.dll SecurityPrototype=true $ARGS

Searching similar issues this seems to be an error related to the AVX support of the machine. If I check the AVX support on the machine with the command

grep avx /proc/cpuinfo

it seems to support AVX but not AVX2

However I executed the same steps on a Windows 10 machine that also supports AVX but not AVX2 and it works fine.

Gabriella Esposito
  • 1

1 Answers1

0

@Gabriella Esposito Azure cognitive services containers require the host to have AVX2 installed or supported on linux systems. Please check the requirements of host computer here.

You can try to enable this and run the container again.

RohitMungi-MSFT
  • 107
  • 5
  • Thanks for the answer. I cheked the AVX requirement on the machine and it suppurts only AVX but not AVX2, is there a way to install this version or is it a pure hardware feature? `[user@machine ~]$ grep -q avx /proc/cpuinfo && echo AVX supported || echo No AVX support detected` `AVX supported` `[user@machine ~]$ grep -q avx2 /proc/cpuinfo && echo AVX2 supported || echo No AVX2 support detected` `No AVX2 support detected` – Gabriella Esposito Jul 29 '20 at 14:22
  • The [release notes from 7.3](https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/7.3_release_notes/new_features_compiler_and_tools) indicate that it would be available on hosts that are capable. My best guess would be your hardware might not be supporting it. – RohitMungi-MSFT Jul 29 '20 at 14:57