NVIDIA-Merlin / Merlin

NVIDIA Merlin is an open source library providing end-to-end GPU-accelerated recommender systems, from feature engineering and preprocessing to training deep learning models and running inference in production.
Apache License 2.0
715 stars 111 forks source link

Allow building Merlin with older versions of `docker`. #1064

Closed bashimao closed 9 months ago

bashimao commented 10 months ago

Modified images according to a trick discovered by @EmmaQiaoCh , which allows compiling the images on ARM machines with old versions of docker and versions that do not come with buildx.

copy-pr-bot[bot] commented 10 months ago

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

github-actions[bot] commented 10 months ago

Documentation preview

https://nvidia-merlin.github.io/Merlin/review/pr-1064

bashimao commented 9 months ago

What this actually fixes:

  1. It fixes the perf_analyzer not being there on ARM64 systems.
  2. If suppresses the error where the Triton fil backend doesn't exist in the base image with older versions of docker.
  3. It fixes a bunch of typos.
  4. I removed the check that prevented installing python3-libnvinfer on ARM64 systems. That was a legacy issue that got resolved through fixes in the upstream image. Since the tritonserver switched to Ubuntu 22.04 as base image, libnvinfer is available.
EmmaQiaoCh commented 9 months ago

/ok to test

EmmaQiaoCh commented 9 months ago

/ok to test

EmmaQiaoCh commented 9 months ago

The copied PR is merged