[CI vehicle, do not merge] config: raw-input ServerArgs follow-up series (#36618-#36622)

[ch-wan]

Motivation

This PR is the CI vehicle for a five-PR series, not a change. Its branch
cheng/gc-ci is one empty placeholder commit on top of PR 5, so a run here
exercises all five commits at once against main. Do not merge it.

The series is the follow-up to the raw-input ServerArgs work (#36250–#36255),
based on d1f14431fdf. Each PR builds on the previous one; review them in order:

1. config: resolution declares, and nothing writes a field #36618 — config: resolution declares, and nothing writes a field
2. config: every handler declares its cuda-graph decisions #36725 — config: every handler declares its cuda-graph decisions (re-opened from config: every handler declares its cuda-graph decisions #36619, whose base branch was deleted by PR 1's merge)
3. config: a parallel leaf with no live counterpart is read bare #36620 — config: a parallel leaf with no live counterpart is read bare
4. config: a parallel size has one spelling; a patched scope declares its own #36621 — config: a parallel size has one spelling; a patched scope declares its own
5. config: the record is not an object that gets passed around #36622 — config: the record is not an object that gets passed around

Together they finish moving resolution's answers off the record: ServerArgs
holds the operator's raw input, and what resolution decided lives in the
declarations and the published config bags. Review comments belong on the
individual PRs above, not here.

Modifications

One empty commit. git diff against PR 5 is empty; git diff against main is
the five PRs, which is the point of the vehicle.

Labels: run-ci, run-ci-extra, bypass-fastfail.

Accuracy Tests

No model-output change in the series: it moves where a configuration value is
read from, not what resolution decides. The equivalent check is a resolution dump
— every field's resolved value for 24 launch shapes, taken in both trees and
compared field by field:

0 differences across 24 shapes × 478 shared fields, against d1f14431fdf.

The one field the series has and the base does not is grpc_worker_threads: on
main it is a public non-field slot assigned in _handle_deprecated_args, and
PR 1 makes it a declared field. Its value is 4 on both sides.

Every guard also runs at each commit of the series, not only at the head — a
PR that is green only on top of its successors is not reviewable on its own. The
set is the config guards plus every registered test the series touches, ~33 files
per boundary, all green.

No GPU accuracy run locally. That gap is what this vehicle exists to close.

Speed Tests and Profiling

No benchmark run, and none is expected to move: nothing in the series changes a
kernel, a schedule, or the shape of any batch. What changes is the source of a
configuration read — a published dataclass attribute instead of a process-group
getter or an accessor hop.

The one place that could have mattered is torch.compile: gate helpers read
parallel leaves inside compiled forwards, and object.__getattribute__
graph-breaks. That was measured rather than assumed — the reads the series
introduces trace under torch.compile(fullgraph=True), pinned by a regression
test.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

This PR is not for merge. It is closed once the five member PRs land.

🤖 Generated with Claude Code

---

CI States

Latest PR Test (Base): ⏳ Run #33110332487
Latest PR Test (Extra): ⏳ Run #33110383631
Latest PR Test (AMD ROCm 7.2): ❌ Run #33110332664

[ch-wan]

@codex

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Already looking forward to the next diff.

Reviewed commit: adaa12d58b

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[ch-wan]

The five member PRs landed on main (#36618, #36725, #36620, #36621, #36622). This vehicle has served its purpose; closing without merge as stated in the description.