[AMD][CI Failure]: Tracking failure for AMD CI Dependencies & Environments

[zhewenl]

Issue to track all AMD CI failures related dependencies & environments:

• (@Alexei-V-Ivanov-AMD) Multi-GPU testing: HW issue is fixed, K8S is killing job randomly(context)
• UV is not enabled in AMD docker (failure)
• torchaudio is not enabled in AMD docker (failure) (Note)
• terratorch is not enabled in AMD docker (Note: conflict with blinker version installed in original docker)
• pqdm, num2words is not enabled in AMD docker
• ....

[github-actions]

CC @hongxiayang @tjtanaa @vllmellm for ROCm-related issue

[tjtanaa]

MAGMA is also not enabled in AMD docker image. The smart resize of llava based model requires torch compiled with MAGMA. Old issues: #25523 and #14914 .

[tjtanaa]

UV is not enabled in AMD docker (failure) - Bugfix PR: #28142

[AndreasKaratzas]

There is also an attempt to enable torchaudio in #28979
Will be adding pqdm and num2words into rocm-test.txt into that PR EOD.

[Alexei-V-Ivanov-AMD]

(@Alexei-V-Ivanov-AMD) Multi-GPU testing: HW issue is fixed, K8S is killing job randomly(context)

The issue has been resolved. Specifically, we've tested the ultimate solution with the following builds:
https://buildkite.com/vllm/amd-ci/builds/1050#019a7f7c-9922-4e2e-b18e-7381b8c3a57e

Please mark this as "complete"/"done" or "closed".

Thank you!

[amdfaa]

UV is not enabled in AMD docker
resolved here: #28142

[amdfaa]

The terrortorch issue has been resolved: #29205