vLLM prefix caching correctness (+ "test") #346
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
meta-cla
bot
added
the
CLA Signed
joecummings
changed the title
JenniferWang
approved these changes
Oct 8, 2025
| According to the vLLM docs (https://docs.vllm.ai/en/v0.9.0/api/vllm/outputs.html#vllm.outputs.RequestOutput), | ||
| this is the number of tokens with a prefix cache hit. So, the logic is that if we run one generation, | ||
| then run another generation with the same start, we should see the number of cached tokens == the length of the prefix. | ||
| logger.info(f"Weight update completed (now v{self.policy_version})") | ||
|
|
||
| @endpoint | ||
| async def _reset_prefix_cache(self): |
There was a problem hiding this comment.
Maybe we can have this convention of
Suggested change
| async def _reset_prefix_cache(self): | |
| async def _test_reset_prefix_cache(self): |
There was a problem hiding this comment.
In this case, I actually don't mind exposing to the end user! It could be used if someone wants to do something custom with their Policy setup.
Jack-Khuu
approved these changes
Oct 8, 2025
| logger.info(f"Weight update completed (now v{self.policy_version})") | ||
|
|
||
| @endpoint | ||
| async def _reset_prefix_cache(self): |
There was a problem hiding this comment.
Another example of needing
meta-pytorch/monarch#1455
| ) | ||
| expected_cached_tokens = 0 | ||
| async for res in vllm_model.generate( | ||
| first_prompt, sampling_params, request_id="first_16" |
There was a problem hiding this comment.
Does the request_id matter here?
There was a problem hiding this comment.
Nah, just for logging.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Context
Are we correctly clearing our prefix cache? Hard to know without proper testing. Hence, this PR.
What changed?
Open questions