Xilinx / inference-server

https://xilinx.github.io/inference-server/
Apache License 2.0
43 stars 13 forks source link

Use VAI 3.0 XRT and XRM #169

Closed varunsh-xilinx closed 1 year ago

varunsh-xilinx commented 1 year ago

Summary of Changes

Closes #112

Motivation

VAI 3.0 is the latest released version and we should stay up-to-date

Implementation

I updated the XRT and XRM packages to match those suggested by the download scripts for VAI 3.0.

There are also some bugs in the VAI runtime that I have to workaround, as noted in the Dockerfile generation script.

Notes

U250 is not officially supported by VAI 3.0 but it still works with this version of XRT and XRM. You will need to continue to use VAI 2.5 xclbins. The existing tests with U250 continue to pass because they use VAI 2.5 models. A later PR will add tests using VAI 3.0 models.

rt-engine has not been updated yet to a VAI 3.0 version because the public Github is out-of-sync. Once those changes are public, then that will also need updating.

This change requires a container rebuild to pull in the updates to VAI.

gbuildx commented 1 year ago

Build successful!