deployment skill: add SGLang cookbook cross-check (analog of recipes.vllm.ai) - #1685
Merged
Conversation
…vllm.ai) The deployment/evaluation skills cross-check vLLM launch commands against recipes.vllm.ai, but the SGLang reference had no equivalent authoritative recipe source. Add the SGLang cookbook (docs.sglang.io/cookbook, the SGLang analog with the same (hw, variant, quant, strategy, nodes) variant-tuple model) as the cross-check source. - SKILL.md: cookbook cross-check note under the SGLang quick-start, mirroring the vLLM cu130/sm_103 block (JS-rendered Mintlify page, raw-markdown fallback in sgl-project/sglang docs_new/, SM120 nightly-image gotcha). - references/sglang.md: authoritative-recipe section, MoE/FP4 backend flag matrix (flashinfer_mxfp4 vs marlin, deepep vs megamoe), and per-hardware notes (Blackwell / Hopper / RTX PRO 6000). Signed-off-by: Zhiyu Cheng <zhiyuc@nvidia.com>
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: 📒 Files selected for processing (1)
📝 WalkthroughWalkthroughThis PR updates SGLang deployment docs to require cross-checking launch flags against the SGLang cookbook (variant selection via URL fragment and raw-markdown retrieval) and adds MoE/FP4 backend flag guidance plus hardware-specific notes (including SM120 nightly image requirement). ChangesSGLang deployment reference and guidance
🎯 2 (Simple) | ⏱️ ~10 minutes 🚥 Pre-merge checks | ✅ 6✅ Passed checks (6 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #1685 +/- ##
==========================================
- Coverage 67.73% 67.66% -0.07%
==========================================
Files 511 511
Lines 56169 56256 +87
==========================================
+ Hits 38044 38066 +22
- Misses 18125 18190 +65
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates the deployment agent skill documentation to treat the SGLang cookbook (docs.sglang.io/cookbook/...) as the authoritative cross-check source for SGLang launch commands, analogous to the existing recipes.vllm.ai cross-check for vLLM.
Changes:
- Adds an SGLang cookbook cross-check callout to the SGLang quick-start section in the deployment skill.
- Expands the SGLang deployment reference with a cookbook “authoritative recipe source” section, MoE/FP4 backend flag matrix, and hardware-specific notes.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.
| File | Description |
|---|---|
.agents/skills/deployment/SKILL.md |
Adds an SGLang cookbook cross-check note under the SGLang quick-start launch command. |
.agents/skills/deployment/references/sglang.md |
Adds cookbook guidance plus backend/hardware notes for SGLang deployments. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+155
to
+156
| > `.../autoregressive/DeepSeek/DeepSeek-V4`), authoritative for parallelism | ||
| > (`--tp` / `--dp` / EP), MoE backends (`--moe-runner-backend flashinfer_mxfp4` |
Comment on lines
+161
to
+162
| > The page is **JS-rendered** — fetch the raw markdown at | ||
| > `github.com/sgl-project/sglang/blob/main/docs_new/cookbook/<category>/<org>/<model>.mdx`. |
Comment on lines
+20
to
+21
| - **Authoritative for:** parallelism layout (`--tp` / `--dp` / EP), MoE backends, | ||
| strategy-driven flags (MTP, CUDA-graph batch sizing), Docker image tag, and the |
Comment on lines
+14
to
+16
| - **JS-rendered (Mintlify).** A plain fetch usually misses the commands. Fetch | ||
| the **raw markdown** instead: | ||
| `github.com/sgl-project/sglang/blob/main/docs_new/cookbook/<category>/<org>/<model>.mdx` |
mxinO
approved these changes
Jun 12, 2026
Comment on lines
+20
to
+21
| - **Authoritative for:** parallelism layout (`--tp` / `--dp` / EP), MoE backends, | ||
| strategy-driven flags (MTP, CUDA-graph batch sizing), Docker image tag, and the |
Apply Copilot review feedback (endorsed by maintainer for the flag-name fix): - Parallelism flags: use --tp / --ep (the flags the reference documents) and refer to multi-node/data-parallel settings generically, instead of the inconsistent --tp / --dp / EP. - Raw-markdown fetch: point at raw.githubusercontent.com/... which serves the .mdx directly, not github.com/.../blob/... which returns rendered HTML. - Drop backslash-escaping of | inside an inline-code table cell; GFM renders the backslashes literally in code spans. Signed-off-by: Zhiyu Cheng <zhiyuc@nvidia.com>
Contributor
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
.agents/skills/deployment/references/sglang.md (1)
104-110:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winFix the
--deepep-moderow formatting.The raw
|characters inauto|normal|low_latencysplit this table into extra columns, so the markdown renders incorrectly and hides part of the guidance. Reformat the row or move the values out of the table.Suggested fix
-| `--deepep-mode auto|normal|low_latency` | DeepEP dispatch mode | +| `--deepep-mode` | DeepEP dispatch mode: `auto`, `normal`, or `low_latency` |🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.agents/skills/deployment/references/sglang.md around lines 104 - 110, The table row for the flag `--deepep-mode` currently contains raw pipe characters ("auto|normal|low_latency") which break the Markdown table; update the `--deepep-mode` row so the cell does not include unescaped pipes — for example, replace the pipes with commas ("auto, normal, low_latency") or wrap each option in inline code/backticks or escape the pipe characters — ensuring the table remains three columns and the description for `--deepep-mode` is fully visible.Source: Linters/SAST tools
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In @.agents/skills/deployment/references/sglang.md:
- Around line 104-110: The table row for the flag `--deepep-mode` currently
contains raw pipe characters ("auto|normal|low_latency") which break the
Markdown table; update the `--deepep-mode` row so the cell does not include
unescaped pipes — for example, replace the pipes with commas ("auto, normal,
low_latency") or wrap each option in inline code/backticks or escape the pipe
characters — ensuring the table remains three columns and the description for
`--deepep-mode` is fully visible.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: 85292ab3-6748-41cb-86b0-bb640bbdcbbf
📒 Files selected for processing (2)
.agents/skills/deployment/SKILL.md.agents/skills/deployment/references/sglang.md
The previous commit dropped the backslash-escaping of | inside the --deepep-mode code span per a Copilot suggestion, but markdownlint MD056 (table-column-count) counts those unescaped pipes as extra column separators (2 cells -> 4) and fails CI. GitHub's renderer tolerates the unescaped form, but the repo's markdownlint does not, so keep the escaping. Verified clean with markdownlint-cli2@0.18.1 (the pinned CI version). Signed-off-by: Zhiyu Cheng <zhiyuc@nvidia.com>
The inline callout duplicated the per-flag/backend detail that now lives in references/sglang.md (the authoritative-recipe section + backend matrix). Drop the flag enumeration from SKILL.md, keeping only the cookbook pointer, URL-fragment variant selection, JS-render/raw-markdown tip, and the SM120 deploy-blocking image gotcha — matching the altitude of the vLLM callout. Detailed parallelism/MoE-backend flags stay in references/sglang.md. Signed-off-by: Zhiyu Cheng <zhiyuc@nvidia.com>
Edwardf0t1
enabled auto-merge (squash)
June 12, 2026 07:42
Contributor
|
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 does this PR do?
Type of change: documentation (agent skill)
The
deploymentandevaluationskills cross-check vLLM launch commands againstrecipes.vllm.ai, but the SGLang reference had no equivalent authoritative recipe source. This adds the SGLang cookbook (docs.sglang.io/cookbook) — the direct SGLang analog ofrecipes.vllm.ai, using the same(hw, variant, quant, strategy, nodes)variant-tuple command generator — as the cross-check source for SGLang deployments.Changes (canonical
.agents/source;.claude/are symlinks):SKILL.md: cookbook cross-check note under the SGLang quick-start, mirroring the existing vLLM cu130/sm_103 block — covers the JS-rendered (Mintlify) page, the raw-markdown fallback insgl-project/sglangdocs_new/, the URL-fragment variant selection, and the SM120 (RTX PRO 6000) nightly-image gotcha.references/sglang.md: a new authoritative-recipe section, a MoE/FP4 backend flag matrix (flashinfer_mxfp4vsmarlin,deepepvsmegamoe), and per-hardware notes (Blackwell / Hopper / RTX PRO 6000).Usage
N/A — documentation only. Agents now cross-check SGLang launch flags against
docs.sglang.io/cookbook/<category>/<org>/<model>, fetching the raw markdown atgithub.com/sgl-project/sglang/blob/main/docs_new/cookbook/<category>/<org>/<model>.mdxwhen the page is JS-rendered.Testing
Markdown-only change to skill docs; no code paths affected.
Before your PR is "Ready for review"
CONTRIBUTING.md: N/A🤖 Generated with Claude Code
Summary by CodeRabbit