feat(pytorch): add PyTorch 2.13 currency scaffold

.claude/rules/ci-architecture.md

@@ -255,7 +255,8 @@ sglang.pr-amzn2023.yml (trigger + gatekeeper + check-changes + discover-configs)
 - PR callers use `cancel-in-progress: true` for build/test jobs (supersede stale runs).
 
 ### Runner fleets
-- `x86-build-runner` — CPU builds (PyTorch, Base, Ray, XGBoost, Lambda)
+- `x86-build-runner` — CPU builds (Base, Ray, XGBoost, Lambda)
+- `x86-pytorch-build-runner` — PyTorch builds (all variants)
 - `x86-vllm-build-runner` — GPU compilation (vLLM, vLLM-Omni)
 - `x86-sglang-build-runner` — GPU compilation (SGLang)
 - `x86-g6xl-runner` — single-GPU tests (1x L4)

.github/config/image/pytorch/2.13-ec2-cpu.yml

@@ -0,0 +1,32 @@
+image:
+  name: "pytorch-2.13-ec2-cpu"
+  description: "PyTorch 2.13 CPU training for EC2 instances"
+
+metadata:
+  framework: "pytorch_runtime"
+  framework_version: "2.13.0"
+  os_version: "amzn2023"
+  customer_type: "ec2"
+  arch_type: "x86"
+  device_type: "cpu"
+  job_type: "training"
+  prod_image: "pytorch:2.13-cpu-amzn2023"
+
+build:
+  dockerfile: "docker/pytorch/Dockerfile.cpu"
+  target: "runtime"
+  python_version: "3.12"
+  torch_version: "2.13.0"
+  torchvision_version: "0.28.0"
+  torchaudio_version: "2.11.0"
+  max_jobs: "8"
+  dlc_major_version: "1"
+  dlc_minor_version: "0"
+
+release:
+  release: true
+  force_release: false
+  public_registry: true
+  private_registry: true
+  enable_soci: true
+  environment: "production"

.github/config/image/pytorch/2.13-ec2-cuda.yml

@@ -0,0 +1,39 @@
+image:
+  name: "pytorch-2.13-ec2-cuda"
+  description: "PyTorch 2.13 CUDA training for EC2 instances"
+
+metadata:
+  framework: "pytorch_runtime"
+  framework_version: "2.13.0"
+  os_version: "amzn2023"
+  customer_type: "ec2"
+  arch_type: "x86"
+  device_type: "gpu"
+  job_type: "training"
+  prod_image: "pytorch:2.13-cu133-amzn2023"
+
+build:
+  dockerfile: "docker/pytorch/Dockerfile.cuda"
+  target: "runtime"
+  python_version: "3.12"
+  cuda_version: "13.3.0"
+  torch_version: "2.13.0"
+  torchvision_version: "0.28.0"
+  torchaudio_version: "2.11.0"
+  flash_attn_version: "2.8.3"
+  deepspeed_version: "0.19.2"
+  transformer_engine_version: "2.17.0"
+  nccl_version: "2.30.7-1"
+  efa_version: "1.49.0"
+  gdrcopy_version: "2.6"
+  max_jobs: "8"
+  dlc_major_version: "1"
+  dlc_minor_version: "0"
+
+release:
+  release: true
+  force_release: false
+  public_registry: true
+  private_registry: true
+  enable_soci: true
+  environment: "production"

.github/config/image/pytorch/2.13-sagemaker-cpu.yml

@@ -0,0 +1,33 @@
+image:
+  name: "pytorch-2.13-sagemaker-cpu"
+  description: "PyTorch 2.13 CPU training for SageMaker"
+
+metadata:
+  framework: "pytorch_runtime"
+  framework_version: "2.13.0"
+  os_version: "amzn2023"
+  customer_type: "sagemaker"
+  platform: "sagemaker"
+  arch_type: "x86"
+  device_type: "cpu"
+  job_type: "training"
+  prod_image: "pytorch:2.13-cpu-amzn2023-sagemaker"
+
+build:
+  dockerfile: "docker/pytorch/Dockerfile.cpu"
+  target: "sagemaker"
+  python_version: "3.12"
+  torch_version: "2.13.0"
+  torchvision_version: "0.28.0"
+  torchaudio_version: "2.11.0"
+  max_jobs: "8"
+  dlc_major_version: "1"
+  dlc_minor_version: "0"
+
+release:
+  release: true
+  force_release: false
+  public_registry: true
+  private_registry: true
+  enable_soci: true
+  environment: "production"

.github/config/image/pytorch/2.13-sagemaker-cuda.yml

@@ -0,0 +1,40 @@
+image:
+  name: "pytorch-2.13-sagemaker-cuda"
+  description: "PyTorch 2.13 CUDA training for SageMaker"
+
+metadata:
+  framework: "pytorch_runtime"
+  framework_version: "2.13.0"
+  os_version: "amzn2023"
+  customer_type: "sagemaker"
+  platform: "sagemaker"
+  arch_type: "x86"
+  device_type: "gpu"
+  job_type: "training"
+  prod_image: "pytorch:2.13-cu133-amzn2023-sagemaker"
+
+build:
+  dockerfile: "docker/pytorch/Dockerfile.cuda"
+  target: "sagemaker"
+  python_version: "3.12"
+  cuda_version: "13.3.0"
+  torch_version: "2.13.0"
+  torchvision_version: "0.28.0"
+  torchaudio_version: "2.11.0"
+  flash_attn_version: "2.8.3"
+  deepspeed_version: "0.19.2"
+  transformer_engine_version: "2.17.0"
+  nccl_version: "2.30.7-1"
+  efa_version: "1.49.0"
+  gdrcopy_version: "2.6"
+  max_jobs: "8"
+  dlc_major_version: "1"
+  dlc_minor_version: "0"
+
+release:
+  release: true
+  force_release: false
+  public_registry: true
+  private_registry: true
+  enable_soci: true
+  environment: "production"

.github/release-schedule.yml

@@ -70,6 +70,10 @@ schedules:
     workflow: vllm-omni.autorelease-ec2.yml
     gpu: medium
 
+  - cron: "30 19 * * 2,4"
+    workflow: pytorch.autorelease-2.13-ec2.yml
+    gpu: light
+
   - cron: "00 20 * * 2,4"
     workflow: sglang.autorelease-ec2-ubuntu.yml
     gpu: heavy
@@ -89,3 +93,7 @@ schedules:
   - cron: "30 22 * * 2"
     workflow: openfold3.autorelease-sagemaker-amzn2023.yml
     gpu: heavy
+
+  - cron: "00 23 * * 2,4"
+    workflow: pytorch.autorelease-2.13-sagemaker.yml
+    gpu: light

.github/workflows/pytorch.autorelease-2.13-ec2.yml

@@ -0,0 +1,40 @@
+name: "Auto Release - PyTorch 2.13 EC2"
+
+on:
+  schedule:
+    - cron: '30 19 * * 2,4'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}
+  cancel-in-progress: false
+
+permissions:
+  contents: read
+
+jobs:
+  discover:
+    runs-on: ubuntu-latest
+    outputs:
+      configs: ${{ steps.discover.outputs.configs }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: discover
+        uses: ./.github/actions/discover-configs
+        with:
+          pattern: ".github/config/image/pytorch/2.13-ec2-*.yml"
+
+  pipeline:
+    needs: [discover]
+    strategy:
+      fail-fast: false
+      matrix:
+        include: ${{ fromJson(needs.discover.outputs.configs) }}
+    uses: ./.github/workflows/pytorch.pipeline.yml
+    with:
+      config-file: ${{ matrix.config_file }}
+      tag-suffix: ${{ github.run_id }}
+      run-multi-gpu-test: false
+      run-multi-node-test: false
+      release: true
+    secrets: inherit

.github/workflows/pytorch.autorelease-2.13-sagemaker.yml

@@ -0,0 +1,40 @@
+name: "Auto Release - PyTorch 2.13 SageMaker"
+
+on:
+  schedule:
+    - cron: '00 23 * * 2,4'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}
+  cancel-in-progress: false
+
+permissions:
+  contents: read
+
+jobs:
+  discover:
+    runs-on: ubuntu-latest
+    outputs:
+      configs: ${{ steps.discover.outputs.configs }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: discover
+        uses: ./.github/actions/discover-configs
+        with:
+          pattern: ".github/config/image/pytorch/2.13-sagemaker-*.yml"
+
+  pipeline:
+    needs: [discover]
+    strategy:
+      fail-fast: false
+      matrix:
+        include: ${{ fromJson(needs.discover.outputs.configs) }}
+    uses: ./.github/workflows/pytorch.pipeline.yml
+    with:
+      config-file: ${{ matrix.config_file }}
+      tag-suffix: ${{ github.run_id }}
+      run-multi-gpu-test: false
+      run-multi-node-test: false
+      release: true
+    secrets: inherit

.github/workflows/pytorch.pipeline.yml

@@ -91,7 +91,7 @@ jobs:
       cancel-in-progress: true
     runs-on:
       - codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }}
-        fleet:x86-build-runner
+        fleet:x86-pytorch-build-runner
         buildspec-override:true
     timeout-minutes: 720
     outputs:

.github/workflows/pytorch.pr-2.13-cpu.yml

@@ -0,0 +1,102 @@
+name: "PR - PyTorch 2.13 CPU"
+
+on:
+  pull_request:
+    branches: [main]
+    types: [opened, reopened, synchronize]
+    paths:
+      - ".github/config/image/pytorch/2.13-*-cpu.yml"
+      - ".github/workflows/pytorch.pipeline.yml"
+      - ".github/workflows/pytorch.pr-2.13-cpu.yml"
+      - "docker/pytorch/Dockerfile.cpu"
+      - "docker/pytorch/2.13/cpu/**"
+      - "scripts/docker/common/**"
+      - "scripts/docker/pytorch/**"
+      - "scripts/docker/telemetry/**"
+      - "test/pytorch/unit/**"
+      - "test/sanity/**"
+      - "test/security/data/ecr_scan_allowlist/pytorch_runtime/**"
+      - "test/telemetry/**"
+      - "!docs/**"
+
+permissions:
+  contents: read
+  pull-requests: read
+
+jobs:
+  gatekeeper:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          ref: ${{ github.event.pull_request.base.sha }}
+          fetch-depth: 1
+      - uses: ./.github/actions/pr-permission-gate
+
+  check-changes:
+    needs: [gatekeeper]
+    runs-on: ubuntu-latest
+    outputs:
+      build-change: ${{ steps.changes.outputs.build-change }}
+      sagemaker-test-change: ${{ steps.changes.outputs.sagemaker-test-change }}
+      sanity-test-change: ${{ steps.changes.outputs.sanity-test-change }}
+      telemetry-test-change: ${{ steps.changes.outputs.telemetry-test-change }}
+      unit-test-change: ${{ steps.changes.outputs.unit-test-change }}
+    steps:
+      - uses: actions/checkout@v6
+      - uses: dorny/paths-filter@v4
+        id: changes
+        with:
+          filters: |
+            build-change:
+              - ".github/config/image/pytorch/2.13-*-cpu.yml"
+              - "docker/pytorch/Dockerfile.cpu"
+              - "docker/pytorch/2.13/cpu/**"
+              - "scripts/docker/common/**"
+              - "scripts/docker/pytorch/**"
+              - "test/security/data/ecr_scan_allowlist/pytorch_runtime/**"
+            sagemaker-test-change:
+              - "test/pytorch/integration/sagemaker/**"
+              - ".github/workflows/pytorch.tests-sagemaker.yml"
+            sanity-test-change:
+              - "test/sanity/**"
+            telemetry-test-change:
+              - "scripts/docker/telemetry/**"
+              - "test/telemetry/**"
+            unit-test-change:
+              - "test/pytorch/unit/**"
+
+  discover:
+    needs: [gatekeeper]
+    runs-on: ubuntu-latest
+    outputs:
+      configs: ${{ steps.discover.outputs.configs }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: discover
+        uses: ./.github/actions/discover-configs
+        with:
+          pattern: ".github/config/image/pytorch/2.13-*-cpu.yml"
+
+  pipeline:
+    needs: [check-changes, discover]
+    strategy:
+      fail-fast: false
+      matrix:
+        include: ${{ fromJson(needs.discover.outputs.configs) }}
+    uses: ./.github/workflows/pytorch.pipeline.yml
+    with:
+      config-file: ${{ matrix.config_file }}
+      tag-suffix: pr-${{ github.event.pull_request.number }}
+      build: ${{ needs.check-changes.outputs.build-change == 'true' }}
+      run-efa-test: false
+      run-multi-gpu-test: false
+      run-multi-node-test: false
+      run-sanity-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.sanity-test-change == 'true' }}
+      run-security-test: ${{ needs.check-changes.outputs.build-change == 'true' }}
+      run-single-gpu-test: false
+      run-sagemaker-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.sagemaker-test-change == 'true' }}
+      run-telemetry-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.telemetry-test-change == 'true' }}
+      run-unit-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.unit-test-change == 'true' }}
+      release: false
+    secrets: inherit