Repository navigation
Conversation
Back-to-back calls hide the NPU's clock drop between requests, which a server sees at its request rate. --gap-ms N sleeps N ms before every timed call (default 0, unchanged).
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
vla-bench --gap-ms Nsleeps N ms before every timed call. The default is 0, so existing runs are unchanged.Why
vla-benchreports the NPU 50–70 ms faster per request than a server actually gets. It also cannot show the effect ofNPU_TURBO.Tests
--gap-ms 0: same timings as before the change.--gap-ms 200on the NPU: SmolVLA 247 ms without turbo vs 178 ms withNPU_TURBO=YES, π0.5 549 vs 487 ms, GR00T N1.6 292 vs 241 ms. That reproduces the clock drop the LIBERO server logs showed.