[build_base] [Docker] Add cuda 11.8 images (#32247)

In order to keep up CUDA versions need for PyTorch 2.0, this PR adds a CUDA 11.8 image. Signed-off-by: Artur Niederfahrenhorst <artur@anyscale.com> Signed-off-by: Kai Fricke <kai@anyscale.com> Co-authored-by: Kai Fricke <kai@anyscale.com>
ray-project · Feb 7, 2023 · 9995599 · 9995599
1 parent ce5a21a
commit 9995599
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 9 deletions.
diff --git a/.buildkite/pipeline.arm64.yml b/.buildkite/pipeline.arm64.yml
@@ -94,7 +94,7 @@
     - pip install -q docker aws_requests_auth boto3
     - ./ci/env/env_info.sh
     - if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then python .buildkite/copy_files.py --destination docker_login; fi
-    - python ./ci/build/build-docker-images.py --py-versions py38 --device-types cu113 cu116 --build-type BUILDKITE --build-base --suffix aarch64
+    - python ./ci/build/build-docker-images.py --py-versions py38 --device-types cu113 cu116 cu118 --build-type BUILDKITE --build-base --suffix aarch64
 
 - label: ":mechanical_arm: :docker: Build Images: py39 [aarch64] (1/2)"
   conditions: ["RAY_CI_LINUX_WHEELS_AFFECTED"]
@@ -114,7 +114,7 @@
     - pip install -q docker aws_requests_auth boto3
     - ./ci/env/env_info.sh
     - if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then python .buildkite/copy_files.py --destination docker_login; fi
-    - python ./ci/build/build-docker-images.py --py-versions py39 --device-types cu113 cu116 --build-type BUILDKITE --build-base --suffix aarch64
+    - python ./ci/build/build-docker-images.py --py-versions py39 --device-types cu113 cu116 cu118 --build-type BUILDKITE --build-base --suffix aarch64
 
 - label: ":mechanical_arm: :docker: Build Images: py310 [aarch64] (1/2)"
   conditions: ["RAY_CI_LINUX_WHEELS_AFFECTED"]
@@ -133,4 +133,4 @@
     - LINUX_WHEELS=1 ./ci/ci.sh build
     - pip install -q docker aws_requests_auth boto3
     - if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then python .buildkite/copy_files.py --destination docker_login; fi
-    - python ./ci/build/build-docker-images.py --py-versions py310 --device-types cu113 cu116 --build-type BUILDKITE --build-base --suffix aarch64
+    - python ./ci/build/build-docker-images.py --py-versions py310 --device-types cu113 cu116 cu118 --build-type BUILDKITE --build-base --suffix aarch64
diff --git a/.buildkite/pipeline.build.yml b/.buildkite/pipeline.build.yml
@@ -122,7 +122,7 @@
     - pip install -q docker aws_requests_auth boto3
     - ./ci/env/env_info.sh
     - if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then python .buildkite/copy_files.py --destination docker_login; fi
-    - python ./ci/build/build-docker-images.py --py-versions py38 --device-types cu111 cu112 cu113 cu116 --build-type BUILDKITE --build-base
+    - python ./ci/build/build-docker-images.py --py-versions py38 --device-types cu111 cu112 cu113 cu116 cu118 --build-type BUILDKITE --build-base
 
 - label: ":docker: Build Images: py39 (1/2)"
   conditions: ["RAY_CI_LINUX_WHEELS_AFFECTED"]
@@ -142,7 +142,7 @@
     - pip install -q docker aws_requests_auth boto3
     - ./ci/env/env_info.sh
     - if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then python .buildkite/copy_files.py --destination docker_login; fi
-    - python ./ci/build/build-docker-images.py --py-versions py39 --device-types cu111 cu112 cu113 cu116 --build-type BUILDKITE --build-base
+    - python ./ci/build/build-docker-images.py --py-versions py39 --device-types cu111 cu112 cu113 cu116 cu118 --build-type BUILDKITE --build-base
 
 - label: ":docker: Build Images: py310 (1/2)"
   conditions: ["RAY_CI_LINUX_WHEELS_AFFECTED"]
@@ -161,7 +161,7 @@
     - LINUX_WHEELS=1 ./ci/ci.sh build
     - pip install -q docker aws_requests_auth boto3
     - if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then python .buildkite/copy_files.py --destination docker_login; fi
-    - python ./ci/build/build-docker-images.py --py-versions py310 --device-types cu111 cu112 cu113 cu116 --build-type BUILDKITE --build-base
+    - python ./ci/build/build-docker-images.py --py-versions py310 --device-types cu111 cu112 cu113 cu116 cu118 --build-type BUILDKITE --build-base
 
 - label: ":java: Java"
   conditions: ["RAY_CI_JAVA_AFFECTED"]

diff --git a/ci/build/build-docker-images.py b/ci/build/build-docker-images.py
@@ -40,6 +40,7 @@
 }
 
 BASE_IMAGES = {
+    "cu118": "nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04",
     "cu116": "nvidia/cuda:11.6.1-cudnn8-devel-ubuntu20.04",
     "cu113": "nvidia/cuda:11.3.1-cudnn8-devel-ubuntu20.04",
     "cu112": "nvidia/cuda:11.2.0-cudnn8-devel-ubuntu20.04",
@@ -52,6 +53,7 @@
 }
 
 CUDA_FULL = {
+    "cu118": "CUDA 11.8",
     "cu116": "CUDA 11.6",
     "cu113": "CUDA 11.3",
     "cu112": "CUDA 11.2",
@@ -63,9 +65,9 @@
 
 # The CUDA version to use for the ML Docker image.
 # If changing the CUDA version in the below line, you should also change the base Docker
-# image being used in ~/ci/docker/Dockerfile.gpu to match the same image being used
+# image being used in ~/ci/docker/Dockerfile.base.gpu to match the same image being used
 # here.
-ML_CUDA_VERSION = "cu116"
+ML_CUDA_VERSION = "cu118"
 
 DEFAULT_PYTHON_VERSION = "py37"
 

diff --git a/ci/docker/base.gpu.Dockerfile b/ci/docker/base.gpu.Dockerfile
@@ -1,4 +1,4 @@
-FROM nvidia/cuda:11.6.1-cudnn8-devel-ubuntu20.04
+FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04
 
 ARG REMOTE_CACHE_URL
 ARG BUILDKITE_PULL_REQUEST

diff --git a/docker/retag-lambda/cuda_versions.txt b/docker/retag-lambda/cuda_versions.txt
@@ -1,3 +1,4 @@
+cu118
 cu116
 cu113
 cu112
-Original file line number
+Diff line change
@@ -1,3 +1,4 @@
+    cu118
     cu116
     cu113
     cu112
@@ Expand Down @@