[CI] Fix mxfp8 TrtllmGenMoe test

[zianglih]

Motivation

@HumansAnd

mxfp8 test regressed after #21667 (reference: #22136).

Modifications

• Use offline quantized checkpoint to avoid online quantization path instability similar to:
  • [CI] [FlashInfer v0.6.7] Use offline quantized checkpoint for MXFP8 Gemm tests #21625
• Use flashinfer_cutlass instead of flashinfer_trtllm for fp8 gemm backend, sinceflashinfer_trtllm seems to require some padding fix for stable numerics, which is missing in current implementation.

Accuracy Tests

Speed Tests and Profiling

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

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[zianglih]

Hi @Fridge003 @b8zhong could you help review this tiny ci fix? Thank you!

[gemini-code-assist]

Code Review

This pull request updates the test configuration for the Flashinfer TRTLLM MoE backend by switching to a specific MXFP8 quantized model and changing the --fp8-gemm-backend to flashinfer_cutlass. Feedback suggests moving the model checkpoint from a personal repository to an official organization account to ensure long-term availability and security for the CI suite. Additionally, there is a concern that switching the GEMM backend to cutlass in a test specifically intended for flashinfer_trtllm reduces test coverage for the target integration, suggesting that the underlying numerical issues should be fixed rather than bypassed.

[b8zhong]

/tag-and-rerun-ci

[zianglih]

/rerun-stage stage-c-test-4-gpu-b200

[zianglih]

/rerun-failed-ci

[zianglih]

/rerun-failed-ci

[zianglih]

/rerun-failed-ci

[zianglih]

/rerun-failed-ci

[zianglih]

/rerun-failed-ci