[Feature]: Improve Logging for Error Messages

[robertgshaw2-redhat]

🚀 The feature, motivation and pitch

Improve logging on VLLM V1 for common errors in initialization.

For example:

• not enough memory to fit the model
• not enough kv cache space to fit the model
• ...

Currently we have decently logging of the exceptions that arise. It would be better however if we could explicitly catch these issues and return clearer error messages

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[13Mai13]

Is the goal to simply raise the same exceptions (like RuntimeError) but with more specific, helpful messages, or should I create custom exception types (like NotEnoughMemoryError, KVCacheSpaceError, etc.) for these issues?

[apehex]

Are you talking about the errors that are already covered or about adding failure modes?
For example, do you want to detail the assert statements?

Do you have an example of what you want and do not want?

@robertgshaw2-redhat @ywang96

[apehex]

I'd be interested to work on this issue as a productive way to browse and understand the codebase.

[mengbingrock]

I'm reviewing V0 for similar logging but didn't see this specific logging at first glance.
Just to clarify, we're not copying the logging from V0 to V1 — instead, we're implementing new logging in V1, correct?

[github-actions]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!

[Sheehan20]

Hi, I'll take this on. Planning to improve error messages for GPU memory, KV cache, and model loading issues.