A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 10, 01:30 UTC
[build] fix cu130 related release pipeline steps and publish as nightly image (#32522)
Signed-off-by: Shengqi Chen <harry-chen@outlook.com>
Shengqi Chen committed
Jan 18, 2026 at 10:36 UTC
965765aef94e4497c67ff30e0c8325e01e16d284
3 files changed
+68
-16
.buildkite/release-pipeline.yaml
+23
-14
@@ -141,7 +141,7 @@ steps:
141
queue: cpu_queue_postmerge
142
commands:
143
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
144
- - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.2 --build-arg FLASHINFER_AOT_COMPILE=true --build-arg INSTALL_KV_CONNECTORS=true --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.2-devel-ubuntu22.04 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130 --target vllm-openai --progress plain -f docker/Dockerfile ."
144
+ - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.1 --build-arg INSTALL_KV_CONNECTORS=true --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.1-devel-ubuntu22.04 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130 --target vllm-openai --progress plain -f docker/Dockerfile ."
145
- "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130"
146
# re-tag to default image tag and push, just in case arm64 build fails
147
- "docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130 public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-cu130"
@@ -154,7 +154,8 @@ steps:
154
queue: arm64_cpu_queue_postmerge
155
commands:
156
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
157
- - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.2 --build-arg FLASHINFER_AOT_COMPILE=true --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0' --build-arg INSTALL_KV_CONNECTORS=true --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.2-devel-ubuntu22.04 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130 --target vllm-openai --progress plain -f docker/Dockerfile ."
157
+ # compute capability 12.0 for RTX-50 series / RTX PRO 6000 Blackwell, 12.1 for DGX Spark
158
+ - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.1 --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0 12.1' --build-arg INSTALL_KV_CONNECTORS=true --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.1-devel-ubuntu22.04 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130 --target vllm-openai --progress plain -f docker/Dockerfile ."
159
- "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)-cu130"
160
161
- label: "Create multi-arch manifest - CUDA 13.0"
@@ -243,7 +244,6 @@ steps:
244
# Build vLLM ROCm image using the base
245
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg BASE_IMAGE=rocm/vllm-dev:base-$BUILDKITE_COMMIT --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-rocm --target vllm-openai --progress plain -f docker/Dockerfile.rocm ."
246
- "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-rocm"
246
-
247
248
- label: "Build and publish nightly multi-arch image to DockerHub"
249
depends_on:
@@ -252,17 +252,7 @@ steps:
252
agents:
253
queue: small_cpu_queue_postmerge
254
commands:
255
- - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
256
- - "docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-x86_64"
257
- - "docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-aarch64"
258
- - "docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-x86_64 vllm/vllm-openai:nightly-x86_64"
259
- - "docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-aarch64 vllm/vllm-openai:nightly-aarch64"
260
- - "docker push vllm/vllm-openai:nightly-x86_64"
261
- - "docker push vllm/vllm-openai:nightly-aarch64"
262
- - "docker manifest create vllm/vllm-openai:nightly vllm/vllm-openai:nightly-x86_64 vllm/vllm-openai:nightly-aarch64 --amend"
263
- - "docker manifest create vllm/vllm-openai:nightly-$BUILDKITE_COMMIT vllm/vllm-openai:nightly-x86_64 vllm/vllm-openai:nightly-aarch64 --amend"
264
- - "docker manifest push vllm/vllm-openai:nightly"
265
- - "docker manifest push vllm/vllm-openai:nightly-$BUILDKITE_COMMIT"
255
+ - "bash .buildkite/scripts/push-nightly-builds.sh"
256
# Clean up old nightly builds (keep only last 14)
257
- "bash .buildkite/scripts/cleanup-nightly-builds.sh"
258
plugins:
@@ -273,6 +263,25 @@ steps:
263
DOCKER_BUILDKIT: "1"
264
DOCKERHUB_USERNAME: "vllmbot"
265
266
+ - label: "Build and publish nightly multi-arch image to DockerHub - CUDA 13.0"
267
+ depends_on:
268
+ - create-multi-arch-manifest-cuda-13-0
269
+ if: build.env("NIGHTLY") == "1"
270
+ agents:
271
+ queue: small_cpu_queue_postmerge
272
+ commands:
273
+ - "bash .buildkite/scripts/push-nightly-builds.sh cu130"
274
+ # Clean up old nightly builds (keep only last 14)
275
+ - "bash .buildkite/scripts/cleanup-nightly-builds.sh cu130-nightly-"
276
+ plugins:
277
+ - docker-login#v3.0.0:
278
+ username: vllmbot
279
+ password-env: DOCKERHUB_TOKEN
280
+ env:
281
+ DOCKER_BUILDKIT: "1"
282
+ DOCKERHUB_USERNAME: "vllmbot"
283
+
284
+
285
# =============================================================================
286
# ROCm Release Pipeline (x86_64 only)
287
# =============================================================================
.buildkite/scripts/cleanup-nightly-builds.sh
+9
-2
@@ -3,7 +3,14 @@
3
set -ex
4
5
# Clean up old nightly builds from DockerHub, keeping only the last 14 builds
6
-# This script uses DockerHub API to list and delete old tags with "nightly-" prefix
6
+# This script uses DockerHub API to list and delete old tags with specified prefix
7
+# Usage: cleanup-nightly-builds.sh [TAG_PREFIX]
8
+# Example: cleanup-nightly-builds.sh "nightly-" or cleanup-nightly-builds.sh "cu130-nightly-"
9
+
10
+# Get tag prefix from argument, default to "nightly-" if not provided
11
+TAG_PREFIX="${1:-nightly-}"
12
+
13
+echo "Cleaning up tags with prefix: $TAG_PREFIX"
14
15
# DockerHub API endpoint for vllm/vllm-openai repository
16
REPO_API_URL="https://hub.docker.com/v2/repositories/vllm/vllm-openai/tags"
@@ -45,7 +52,7 @@ get_all_tags() {
52
set -x
53
54
# Get both last_updated timestamp and tag name, separated by |
48
- local tags=$(echo "$response" | jq -r '.results[] | select(.name | startswith("nightly-")) | "\(.last_updated)|\(.name)"')
55
+ local tags=$(echo "$response" | jq -r --arg prefix "$TAG_PREFIX" '.results[] | select(.name | startswith($prefix)) | "\(.last_updated)|\(.name)"')
56
57
if [ -z "$tags" ]; then
58
break
.buildkite/scripts/push-nightly-builds.sh
new
+36
@@ -0,0 +1,36 @@
1
+#!/bin/bash
2
+
3
+set -ex
4
+
5
+# Get tag variant from argument, default to empty if not provided, should be something like "cu130".
6
+# Due to limits in cleanup script, we must move variants to use separate tags like "cu130-nightly",
7
+# otherwise they will be cleaned up together with the main "nightly" tags.
8
+
9
+TAG_VARIANT="$1"
10
+if [ -n "$TAG_VARIANT" ]; then
11
+ ORIG_TAG_SUFFIX="-$TAG_VARIANT"
12
+ TAG_NAME="$TAG_VARIANT-nightly"
13
+else
14
+ ORIG_TAG_SUFFIX=""
15
+ TAG_NAME="nightly"
16
+fi
17
+
18
+ORIG_TAG_NAME="$BUILDKITE_COMMIT"
19
+
20
+echo "Pushing original tag $ORIG_TAG_NAME$ORIG_TAG_SUFFIX to new nightly tag name: $TAG_NAME"
21
+
22
+# pull original arch-dependent images from AWS ECR Public
23
+aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7
24
+docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:$ORIG_TAG_NAME-x86_64$ORIG_TAG_SUFFIX
25
+docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:$ORIG_TAG_NAME-aarch64$ORIG_TAG_SUFFIX
26
+# tag arch-dependent images
27
+docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$ORIG_TAG_NAME-x86_64$ORIG_TAG_SUFFIX vllm/vllm-openai:$TAG_NAME-x86_64
28
+docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$ORIG_TAG_NAME-aarch64$ORIG_TAG_SUFFIX vllm/vllm-openai:$TAG_NAME-aarch64
29
+# push arch-dependent images to DockerHub
30
+docker push vllm/vllm-openai:$TAG_NAME-x86_64
31
+docker push vllm/vllm-openai:$TAG_NAME-aarch64
32
+# push arch-independent manifest to DockerHub
33
+docker manifest create vllm/vllm-openai:$TAG_NAME vllm/vllm-openai:$TAG_NAME-x86_64 vllm/vllm-openai:$TAG_NAME-aarch64 --amend
34
+docker manifest create vllm/vllm-openai:$TAG_NAME-$BUILDKITE_COMMIT vllm/vllm-openai:$TAG_NAME-x86_64 vllm/vllm-openai:$TAG_NAME-aarch64 --amend
35
+docker manifest push vllm/vllm-openai:$TAG_NAME
36
+docker manifest push vllm/vllm-openai:$TAG_NAME-$BUILDKITE_COMMIT