Skip to content

Fix config for 1k/8k#94

Merged
ishandhanani merged 2 commits intomainfrom
kylliang/fix_container_mismatch
Jan 23, 2026
Merged

Fix config for 1k/8k#94
ishandhanani merged 2 commits intomainfrom
kylliang/fix_container_mismatch

Conversation

@kyleliang-nv
Copy link
Copy Markdown
Collaborator

@kyleliang-nv kyleliang-nv commented Jan 23, 2026

Be exact on which container to use.

Summary by CodeRabbit

  • Chores
    • Updated container image versions to the latest nightly build for improved compatibility and stability.
    • Refactored FP4 GEMM backend configuration for optimized performance across multiple deployment scenarios.

✏️ Tip: You can customize this high-level summary in your review settings.

@coderabbitai
Copy link
Copy Markdown
Contributor

coderabbitai bot commented Jan 23, 2026

📝 Walkthrough

Walkthrough

Three GB200 FP4 1k/8k recipe YAML configuration files are updated with a new container image version and FP4 GEMM backend settings are migrated from environment variables to sglang_config parameters.

Changes

Cohort / File(s) Summary
Container image and FP4 GEMM backend configuration updates
recipies/gb200-fp4/1k8k/low-latency.yaml
Container image updated to nightly-dev-cu13-20260121-1e309030
Container image and FP4 GEMM backend configuration updates
recipies/gb200-fp4/1k8k/max-tpt.yaml
Container image updated; SGLANG_FLASHINFER_FP4_GEMM_BACKEND removed from prefill environment; fp4-gemm-backend: "flashinfer_cutlass" added to prefill sglang_config
Container image and FP4 GEMM backend configuration updates
recipies/gb200-fp4/1k8k/mid-curve.yaml
Container image updated; SGLANG_FLASHINFER_FP4_GEMM_BACKEND removed from decode environment; fp4-gemm-backend: "flashinfer_cutlass" added to both prefill and decode sglang_config

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Possibly related PRs

  • #85: Updates the same GB200 FP4 recipe files with container image and FP4 GEMM backend configuration changes.

Suggested reviewers

  • ishandhanani

Poem

🐰 Hop hop, images renewed!
From dev to nightly, our configs brewed,
Env vars transformed to config grace,
FP4 gemm settings find their place!

🚥 Pre-merge checks | ✅ 3
✅ Passed checks (3 passed)
Check name Status Explanation
Description Check ✅ Passed Check skipped - CodeRabbit’s high-level summary is enabled.
Title check ✅ Passed The title 'Fix config for 1k/8k' directly relates to the changeset, which updates configuration files in the recipies/gb200-fp4/1k8k/ directory with container image updates and configuration adjustments.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✏️ Tip: You can configure your own custom pre-merge checks in the settings.


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands and usage tips.

Copy link
Copy Markdown
Contributor

@coderabbitai coderabbitai bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actionable comments posted: 0

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)
recipies/gb200-fp4/1k8k/max-tpt.yaml (1)

124-124: Potential typo: flashinfer_cutedsl vs flashinfer_cutlass.

The decode section uses moe-runner-backend: "flashinfer_cutedsl" while the prefill section (line 73) uses moe-runner-backend: "flashinfer_cutlass". If cutedsl is a valid distinct backend for decode, please ignore. Otherwise, this may be a typo that should be corrected.

recipies/gb200-fp4/1k8k/mid-curve.yaml (1)

125-125: Potential typo: flashinfer_cutedsl vs flashinfer_cutlass.

Same as max-tpt.yaml—decode uses "flashinfer_cutedsl" while prefill (line 74) uses "flashinfer_cutlass". Please verify this is intentional.

@ishandhanani ishandhanani merged commit cd945f4 into main Jan 23, 2026
4 of 5 checks passed
@Fridge003
Copy link
Copy Markdown
Collaborator

@kyleliang-nv Can we change the containers to 0.3.8 later? The nightly images will be automatically deleted after 15 days

@kyleliang-nv
Copy link
Copy Markdown
Collaborator Author

@kyleliang-nv Can we change the containers to 0.3.8 later? The nightly images will be automatically deleted after 15 days

Yes, I am aware that nightly containers are purged after certain days. I'm using the nightly container as a way to pin to an exact container until the 0.5.8 is released, then I'll switch the container to that one

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants