From 478b4b2579ee3ec1c65a66b49412c6e3aded3775 Mon Sep 17 00:00:00 2001 From: gitlawr Date: Thu, 21 May 2026 15:19:42 +0800 Subject: [PATCH 1/2] refactor(cuda/rocm): bump vllm to 0.20.2 Signed-off-by: gitlawr --- pack/cuda/Dockerfile.vllm | 4 ++-- pack/matrix.yaml | 12 ++++++------ pack/rocm/Dockerfile.vllm | 4 ++-- 3 files changed, 10 insertions(+), 10 deletions(-) diff --git a/pack/cuda/Dockerfile.vllm b/pack/cuda/Dockerfile.vllm index 2848823..0410940 100644 --- a/pack/cuda/Dockerfile.vllm +++ b/pack/cuda/Dockerfile.vllm @@ -2,8 +2,8 @@ ARG PYTHON_VERSION=3.12 ARG CMAKE_MAX_JOBS ARG CUDA_VERSION=13.0.1 ARG CUDA_ARCHS -ARG VLLM_BASE_IMAGE=vllm/vllm-openai:v0.20.1 -ARG VLLM_VERSION=0.20.1 +ARG VLLM_BASE_IMAGE=vllm/vllm-openai:v0.20.2 +ARG VLLM_VERSION=0.20.2 ARG VLLM_TORCH_VERSION=2.12.0 ARG VLLM_TORCH_CUDA_VERSION=${CUDA_VERSION} ARG VLLM_OMNI_COMMIT=704b6bb6 diff --git a/pack/matrix.yaml b/pack/matrix.yaml index 2107fc9..a4f0c9d 100644 --- a/pack/matrix.yaml +++ b/pack/matrix.yaml @@ -60,8 +60,8 @@ rules: - "linux/amd64" args: - "ROCM_VERSION=7.2.1" - - "VLLM_VERSION=0.20.1" - - "VLLM_BASE_IMAGE=vllm/vllm-openai-rocm:v0.20.1" + - "VLLM_VERSION=0.20.2" + - "VLLM_BASE_IMAGE=vllm/vllm-openai-rocm:v0.20.2" # # NVIDIA CUDA @@ -74,8 +74,8 @@ rules: - "vllm" args: - "CUDA_VERSION=13.0.1" - - "VLLM_VERSION=0.20.1" - - "VLLM_BASE_IMAGE=vllm/vllm-openai:v0.20.1-ubuntu2404" + - "VLLM_VERSION=0.20.2" + - "VLLM_BASE_IMAGE=vllm/vllm-openai:v0.20.2-ubuntu2404" ## NVIDIA CUDA 12.9.1 ## - backend: "cuda" @@ -83,8 +83,8 @@ rules: - "vllm" args: - "CUDA_VERSION=12.9.1" - - "VLLM_VERSION=0.20.1" - - "VLLM_BASE_IMAGE=vllm/vllm-openai:v0.20.1-cu129-ubuntu2404" + - "VLLM_VERSION=0.20.2" + - "VLLM_BASE_IMAGE=vllm/vllm-openai:v0.20.2-cu129-ubuntu2404" # # Iluvatar CoreX diff --git a/pack/rocm/Dockerfile.vllm b/pack/rocm/Dockerfile.vllm index a666e53..bdf4eab 100644 --- a/pack/rocm/Dockerfile.vllm +++ b/pack/rocm/Dockerfile.vllm @@ -2,8 +2,8 @@ ARG PYTHON_VERSION=3.12 ARG CMAKE_MAX_JOBS ARG ROCM_VERSION=7.2.1 ARG ROCM_ARCHS -ARG VLLM_BASE_IMAGE=vllm/vllm-openai-rocm:v0.20.1 -ARG VLLM_VERSION=0.20.1 +ARG VLLM_BASE_IMAGE=vllm/vllm-openai-rocm:v0.20.2 +ARG VLLM_VERSION=0.20.2 ARG VLLM_TORCH_VERSION=2.10.0 ARG VLLM_TORCH_ROCM_VERSION=${ROCM_VERSION} ARG VLLM_LMCACHE_VERSION=0.4.4 From 86022f234c4f972e945442c1ef0cdab5e691981d Mon Sep 17 00:00:00 2001 From: gitlawr Date: Thu, 21 May 2026 16:13:54 +0800 Subject: [PATCH 2/2] fix(make): honor service-specific Dockerfile in package target Match the workflow behavior in .github/workflows/pack.yml: when pack//Dockerfile. exists, use it instead of the backend's main Dockerfile. Signed-off-by: gitlawr --- Makefile | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/Makefile b/Makefile index 1704cc8..0e4956a 100644 --- a/Makefile +++ b/Makefile @@ -142,7 +142,7 @@ package: --platform "$${JOB_PLATFORM}" \ --target "$${JOB_TARGET}" \ --tag "$${JOB_TAG}" \ - --file "$${JOB_LOCATION}/Dockerfile" \ + --file "$${JOB_DOCKERFILE}" \ --attest "type=provenance,disabled=true" \ --attest "type=sbom,disabled=true" \ --ulimit nofile=65536:65536 \