vllm

[Bugfix] Use is_integrated to detect UMA GPUs for memory reporting

#35356

Merged