-
-
Notifications
You must be signed in to change notification settings - Fork 11.4k
[WIP][KVConnector] Retrive KV events from LMCache #28252
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
Updates to the LMCache connector to enable events generated by LMCache to be retrieved and published by the scheduler. Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Code Review
This pull request introduces functionality to retrieve KV events from LMCache by adding a new LMCacheKVEvents class and integrating it into the LMCacheConnectorV1. This is a good step towards better KV-cache interoperability. However, the implementation of LMCacheKVEvents contains several critical bugs in the aggregate, is_empty, and reduce methods that would prevent it from working as intended. My review includes specific code suggestions to fix these issues and ensure the new functionality is robust and correct.
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py
Outdated
Show resolved
Hide resolved
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py
Outdated
Show resolved
Hide resolved
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py
Outdated
Show resolved
Hide resolved
|
My initial reaction is that if a connector needs to send events from the worker-side to the scheduler-side, we should add events to |
I agree @markmc, thanks. This is my first effort by piggybacking on the stats. I will push a separate PR which adds events to KVConnectorOutput first and then rebase this accordingly.. |
This commit follows recommendation from @markmc to use a specific event property instead of piggybacking on stats. PR vllm-project#28309 adds the events property to KVConnectorOutput and this commit picks up the new property and uses it to pass the events from worker side to scheduler side. Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
|
This pull request has merge conflicts that must be resolved before it can be |
Purpose
Updates to the LMCache connector to enable KV events generated by LMCache to be retrieved and published by the scheduler.
This is part of the solution for:
Note to reviewers: Do NOT merge until the following merge first:
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.