Disable prefix caching + chunked prefill in spec_compression_stress.sh#35
Merged
Conversation
The simulator defaults --enable-prefix-caching and --enable-chunked-prefill to true. The stress script never opted out, so all three modes (baseline / self_verify / cpu_verify) were implicitly comparing radix-attention + chunked-prefill paths instead of the lean prefill/decode pipeline the comparison is designed to isolate. Pin both off so the run results reflect only the spec-decode and KV-compression deltas the stress is meant to study. Print the choice in the run banner so it's visible in the log.
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Pins
--no-enable-prefix-cachingand--no-enable-chunked-prefillfor every mode inserving/spec_compression_stress.sh.Why
The simulator's CLI defines both as
BooleanOptionalActionwithdefault=True(serving/__main__.py:112,115). The stress script never opted out, so all three modes — baseline, self_verify, cpu_verify — were implicitly running with radix-attention prefix reuse + chunked-prefill. That's not what the comparison is designed to isolate; the stress is meant to measure how spec-decode and periodic KV compression interact on a lean prefill / decode pipeline, and leaving prefix-caching + chunked-prefill on confounds the result.Change
Two lines added to the
run_modecmd block so every dispatched mode inherits the flags:Plus one banner line so the choice shows up in the run log alongside the workload / spec / compression summary.
Test plan
./serving/spec_compression_stress.sh baseline(or./...no-arg) printsScheduler: prefix-caching=off chunked-prefill=offin the banner.Generated by Claude Code