OpenHands is also super helpful for me to figure out OOM issues from vLLM 😄
This issue has bugged me for a while, but I'm too unfamiliar with the vllm codebase and don't really have the patience to go through the code myself -- so I have to live with the pod dying every 15 the minutes, getting restarted, and die again.
I thought maybe i could throw the stack traces and have OpenHands help me figure out stuff -- and it gave me a usable solution in two minutes, and it actually worked!