fix: adding teacache.params back to sampling params as intended

[dougyster]

Motivation

Currently, adding enable_teacache=true fails with 'Req' object has no attribute 'teacache_params'. This bug was introduced in PR #19964 which intended to remove teacache_params from the request body so fallback to sampling params to be handled, but teacache_params were not on sampling params.

Also added teacache params for Wan 2.2 using the Wan2.1 14B coefficients as a proxy so that enable_teacache actually takes effect.

Modifications

Adding teacache_params back to sampling params and added teacache_params: to class Wan2_2_Base_SamplingParams.

Accuracy Tests

python3 -m
  sglang.multimodal_gen.runtime.entrypoints.cli.main
  serve \
    --model-path Wan-AI/Wan2.2-T2V-A14B-Diffusers \
    --dit-layerwise-offload \
    --host 0.0.0.0 --port 8080

  Request:
  curl "http://localhost:8080/v1/videos" \
    -F "prompt=A dog running through a park" \
    -F "size=832x480" \
    -F "num_frames=81" \
    -F "fps=24" \
    -F "num_inference_steps=50" \
    -F "guidance_scale=5.0" \
    -F "negative_prompt=blurry, low quality, distorted"
   \
    -F "seed=42" \
    -F "enable_teacache=true"

wan22-teacache-test.mp4

Benchmarking and Profiling

Checklist

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

Review Process

1. Ping Merge Oncalls to start the PR flow. 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.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[mickqian]

could we modify a testcase to guard this model?

[dougyster]

could we modify a testcase to guard this model?

Added a teacache test for Wan2.2!

[mickqian]

/tag-and-rerun-ci

[yhyang201]

/tag-and-rerun-ci

[yhyang201]

@mickqian All CI (Nvidia + AMD) passed and PR is approved, ready for merge

— SGLDHelper bot