[PyTorch][ARM64][Training][EC2] PT 2.9.0 Release

dlc_developer_config.toml

@@ -15,7 +15,7 @@ neuronx_mode = false
 graviton_mode = false
 # Please only set it to true if you are preparing a ARM64 related PR
 # Do remember to revert it back to false before merging any PR (including ARM64 dedicated PR)
-arm64_mode = false
+arm64_mode = true
 # Please only set it to True if you are preparing a HABANA related PR
 # Do remember to revert it back to False before merging any PR (including HABANA dedicated PR)
 habana_mode = false
@@ -37,12 +37,12 @@ deep_canary_mode = false
 [build]
 # Add in frameworks you would like to build. By default, builds are disabled unless you specify building an image.
 # available frameworks - ["base", "vllm", "sglang", "autogluon", "huggingface_tensorflow", "huggingface_pytorch", "huggingface_tensorflow_trcomp", "huggingface_pytorch_trcomp", "pytorch_trcomp", "tensorflow", "pytorch", "stabilityai_pytorch"]
-build_frameworks = []
+build_frameworks = ["pytorch"]
 
 
 # By default we build both training and inference containers. Set true/false values to determine which to build.
 build_training = true
-build_inference = true
+build_inference = false
 
 # Set do_build to "false" to skip builds and test the latest image built by this PR
 # Note: at least one build is required to set do_build to "false"
@@ -69,7 +69,7 @@ ecs_tests = true
 eks_tests = true
 ec2_tests = true
 # Set it to true if you are preparing a Benchmark related PR
-ec2_benchmark_tests = false
+ec2_benchmark_tests = true
 
 ### Set ec2_tests_on_heavy_instances = true to be able to run any EC2 tests that use large/expensive instance types by
 ### default. If false, these types of tests will be skipped while other tests will run as usual.
@@ -78,7 +78,7 @@ ec2_benchmark_tests = false
 ec2_tests_on_heavy_instances = false
 ### SM specific tests
 ### On by default
-sagemaker_local_tests = true
+sagemaker_local_tests = false
 ### Set enable_ipv6 = true to run tests with IPv6-enabled resources
 ### Off by default (set to false)
 enable_ipv6 = false
@@ -96,7 +96,7 @@ enable_ipv6 = false
 ipv6_vpc_name = ""
 
 # run standard sagemaker remote tests from test/sagemaker_tests
-sagemaker_remote_tests = true
+sagemaker_remote_tests = false
 # run efa sagemaker tests
 sagemaker_efa_tests = false
 # run release_candidate_integration tests
@@ -129,7 +129,7 @@ dlc-pr-tensorflow-2-training = ""
 dlc-pr-autogluon-training = ""
 
 # ARM64 Training
-dlc-pr-pytorch-arm64-training = ""
+dlc-pr-pytorch-arm64-training = "pytorch/training/buildspec-arm64-2-9-ec2.yml"
 
 # HuggingFace Training
 dlc-pr-huggingface-tensorflow-training = ""

pytorch/training/buildspec-arm64-2-9-ec2.yml

@@ -0,0 +1,57 @@
+account_id: &ACCOUNT_ID <set-$ACCOUNT_ID-in-environment>
+prod_account_id: &PROD_ACCOUNT_ID 763104351884
+region: &REGION <set-$REGION-in-environment>
+framework: &FRAMEWORK pytorch
+version: &VERSION 2.9.0
+short_version: &SHORT_VERSION "2.9"
+arch_type: arm64
+#autopatch_build: "True"
+
+repository_info:
+  training_repository: &TRAINING_REPOSITORY
+    image_type: &TRAINING_IMAGE_TYPE training
+    root: !join [ *FRAMEWORK, "/", *TRAINING_IMAGE_TYPE ]
+    repository_name: &REPOSITORY_NAME !join [ pr, "-", *FRAMEWORK, "-", *TRAINING_IMAGE_TYPE, "-", arm64 ]
+    repository: &REPOSITORY !join [ *ACCOUNT_ID, .dkr.ecr., *REGION, .amazonaws.com/, *REPOSITORY_NAME ]
+    release_repository_name: &RELEASE_REPOSITORY_NAME !join [ *FRAMEWORK, "-", *TRAINING_IMAGE_TYPE, "-", arm64 ]
+    release_repository: &RELEASE_REPOSITORY !join [ *PROD_ACCOUNT_ID, .dkr.ecr., *REGION, .amazonaws.com/, *RELEASE_REPOSITORY_NAME ]
+
+context:
+  training_context: &TRAINING_CONTEXT
+    start_cuda_compat:
+      source: docker/build_artifacts/start_cuda_compat.sh
+      target: start_cuda_compat.sh
+    dockerd_entrypoint:
+      source: docker/build_artifacts/dockerd_entrypoint.sh
+      target: dockerd_entrypoint.sh
+    changehostname:
+      source: docker/build_artifacts/changehostname.c
+      target: changehostname.c
+    start_with_right_hostname:
+      source: docker/build_artifacts/start_with_right_hostname.sh
+      target: start_with_right_hostname.sh
+    example_mnist_file:
+      source: docker/build_artifacts/mnist.py
+      target: mnist.py
+    deep_learning_container:
+      source: ../../src/deep_learning_container.py
+      target: deep_learning_container.py
+
+images:
+  BuildEC2Arm64GPUPTTrainPy3cu128DockerImage:
+    <<: *TRAINING_REPOSITORY
+    build: &PYTORCH_GPU_TRAINING_PY3 false
+    image_size_baseline: 19700
+    device_type: &DEVICE_TYPE gpu
+    python_version: &DOCKER_PYTHON_VERSION py3
+    tag_python_version: &TAG_PYTHON_VERSION py312
+    cuda_version: &CUDA_VERSION cu130
+    os_version: &OS_VERSION ubuntu22.04
+    tag: !join [ *VERSION, "-", *DEVICE_TYPE, "-", *TAG_PYTHON_VERSION, "-", *CUDA_VERSION, "-", *OS_VERSION, "-ec2" ]
+    latest_release_tag: !join [ *VERSION, "-", *DEVICE_TYPE, "-", *TAG_PYTHON_VERSION, "-", *CUDA_VERSION, "-", *OS_VERSION, "-ec2" ]
+    # skip_build: "False"
+    docker_file: !join [ docker/, *SHORT_VERSION, /, *DOCKER_PYTHON_VERSION, /, *CUDA_VERSION, /Dockerfile.arm64.,
+                         *DEVICE_TYPE ]
+    target: ec2
+    context:
+      <<: *TRAINING_CONTEXT

pytorch/training/buildspec-arm64.yml

@@ -1 +1 @@
-buildspec_pointer: buildspec-arm64-2-7-ec2.yml
+buildspec_pointer: buildspec-arm64-2-9-ec2.yml