feat: add vLLM-Omni EC2 and SageMaker DLC images#5868

Merged

Yadan-Wei merged 65 commits intomainfrom

Apr 7, 2026

Commits on Apr 2, 2026

feat: add vLLM-Omni EC2 and SageMaker DLC images
Yadan Wei
committed
fix: use AL2023-compatible packages for omni system deps
Yadan Wei
committed
fix: only install ffmpeg static binary for omni deps
Yadan Wei
committed
fix: use SPAL repo for espeak-ng, sox, ffmpeg on AL2023
Yadan Wei
committed
fix: use --region instead of --aws-region for pytest
Yadan Wei
committed
fix: add sagemaker SDK dep and match existing test pattern
Yadan Wei
committed
fix: increase stage init timeout for omni model tests
Yadan Wei
committed
fix: use download-model action for model downloads
Yadan Wei
committed
fix: patch CVE-2026-28414 gradio path traversal in omni image
Yadan Wei
committed
fix: use .tar.gz model tarballs for download-model action compatibility
Yadan Wei
committed
fix: use /v1/audio/speech API for TTS smoke test
Yadan Wei
committed
fix: use HuggingFace model IDs directly instead of S3 tarballs
Yadan Wei
committed
fix: validate diffusion response without printing full base64 image
Yadan Wei
committed
fix: use ml.g4dn.xlarge for TTS endpoint test (cheaper, 1.7B fits in 16GB T4)
Yadan Wei
committed
fix: remove redundant --enforce-eager (vllm-omni enforces it internally)
Yadan Wei
committed
fix: use customer-type from config to select smoke test script
Yadan Wei
committed

Commits on Apr 3, 2026

fix lmiv22 yml and add lmiv23 (#5869)

smouaa
authored and
Yadan Wei
committed
fix telemetry ingress rules (#5871)

sirutBuasai
authored and
Yadan Wei
committed
Migrate Xgboost Container Tests to DLC repo (#5860)

Jyothirmaikottu
authored and
Yadan Wei
committed
fix: use download-model action and /models/ path for omni smoke tests
Yadan Wei
committed
Merge branch 'main' into omni
Yadan-Wei
authored
ci: trigger pipeline
Yadan Wei
committed
Merge branch 'main' into omni
Yadan-Wei
authored
ci: re-trigger after flux2 model tarball fix
Yadan Wei
committed
fix: SM endpoint test validates deployment only (TTS uses /v1/audio/speech, not /invocations)
Yadan Wei
committed
Revert "fix: SM endpoint test validates deployment only (TTS uses /v1/audio/speech, not /invocations)"
Yadan Wei
committed
ci: Disable all non-omni PR workflows
Yadan Wei
committed
feat: add SageMaker serve proxy to route /invocations to correct vllm-omni endpoint
Yadan Wei
committed
feat: SageMaker routing middleware, real entrypoint smoke tests, unit tests
Yadan Wei
committed
feat: pre-built runtime base to skip vLLM compile in PR builds
Yadan Wei
committed
feat: per-model test config with route/request/validate, g5 for endpoint test
Yadan Wei
committed

Commits on Apr 4, 2026

Commits on Apr 6, 2026

feat: move unit test to test/vllm-omni/sagemaker/, add async endpoint test
Yadan Wei
committed
fix: run unit test from sagemaker dir to avoid test/__init__.py import
Yadan Wei
committed
fix: use default-runner for unit test (has test_utils and starlette)
Yadan Wei
committed
fix: install test deps and set PYTHONPATH for unit test (matches sanity pattern)
Yadan Wei
committed
fix: add starlette to unit test deps (not in test/requirements.txt)
Yadan Wei
committed
feat: add 4 new models (CosyVoice3, Qwen2.5-Omni, BAGEL, Wan2.1), HF model support, consolidate tests
Yadan Wei
committed
fix: revert to S3-cached models only, new HF models need validation first
Yadan Wei
committed
Merge branch 'main' into omni
Yadan-Wei
authored
Merge branch 'main' into omni
Yadan-Wei
authored
feat: add CosyVoice3-0.5B and Qwen2.5-Omni-3B smoke tests (S3 cached)
Yadan Wei
committed
fix: bump new models to g6exl (more RAM), add container log dump on failure
Yadan Wei
committed
fix: revert to Qwen3-TTS and FLUX.2 only
Yadan Wei
committed
feat: add CosyVoice3, Wan2.1, BAGEL, Qwen2.5-Omni smoke tests
Yadan Wei
committed
change instance type
Yadan Wei
committed
fix: use absolute path for cosyvoice3 stage config in DLC container
Yadan Wei
committed
fix path
Yadan Wei
committed
feat: add 4 new models, form data support, endpoint cleanup, more logs
Yadan Wei
committed
fix: remove CosyVoice3 - transformers doesn't recognize cosyvoice3 model_type
Yadan Wei
committed
fix: use bash array for curl form data to preserve header quoting
Yadan Wei
committed
fix: Wan2.1 use /v1/videos (async), /v1/videos/sync not in v0.18.0
Yadan Wei
committed

Commits on Apr 7, 2026