Repository navigation
Conversation
knukiban
requested review from
ByronHsu,
Ying1123,
hnyls2002,
ispobock,
merrymercy and
zhyncs
as code owners
March 20, 2025 01:16
Contributor
|
This pull request has been automatically closed due to inactivity. Please feel free to reopen it if needed. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Motivation
ServerArgs.__post_init__is not idempotent and is called twice when runningbench_offline_throughput, causingchunked_prefill_sizeto be halved multiple times.Additionally, https://github.com/sgl-project/sglang/pull/3964/files#diff-700b5118b493d60d7b5994857f5f1e6a7e842ad702392b8ab199945764dfc8edR1144 caused a port collision. (fixed in #4648 though less robustly)
Modifications
--chunked-prefill-sizeor--schedule-conservativenessvalue if explicitly set. Defaults when runningsglang.launch_serverare unchanged frommain.ServerArgs.__post_init__()idempotent, so that runningsglang.bench_offline_throughputuses the same values assglang.launch_serverPortArgswhen using--enable_dp_attentionResult: This completes and uses default chunked prefill size 4096 on 2 GPUs.
Checklist