Merge branch 'main' into es/extras-xformers

# Conflicts:
#	torch-extras/Dockerfile

.github/configurations/torch-base.yml

@@ -0,0 +1,5 @@
+cuda: [ 12.2.2, 12.1.1, 12.0.1, 11.8.0 ]
+include:
+  - torch: 2.0.1
+    vision: 0.15.2
+    audio: 2.0.2

.github/configurations/torch-nccl.yml

@@ -0,0 +1,17 @@
+image:
+  - cuda: 12.2.2
+    nccl: 2.18.5-1
+    nccl-tests-hash: a6a61ab
+  - cuda: 12.1.1
+    nccl: 2.18.3-1
+    nccl-tests-hash: 253a5b1
+  - cuda: 12.0.1
+    nccl: 2.18.5-1
+    nccl-tests-hash: a6a61ab
+  - cuda: 11.8.0
+    nccl: 2.16.2-1
+    nccl-tests-hash: a6a61ab
+include:
+  - torch: 2.0.1
+    vision: 0.15.2
+    audio: 2.0.2

.github/workflows/bloom.yml

@@ -10,7 +10,8 @@ on:
 jobs:
   build:
     uses: ./.github/workflows/build.yml
-    with: 
+    secrets: inherit
+    with:
       image-name: bloom
       folder: bloom
-      build-args: ""
+      build-args: ""

.github/workflows/build.yml

@@ -29,7 +29,7 @@ on:
 jobs:
   build:
     name: Build Images
-    runs-on: [self-hosted, Linux]
+    runs-on: [ self-hosted, Linux ]
     outputs:
       outcome: ${{ steps.docker-build.outcome }}
       tags: ${{ steps.meta.outputs.tags }}
@@ -38,12 +38,17 @@ jobs:
       - uses: actions/checkout@v3
       - name: Set up Docker Buildx
         uses: docker/[email protected]
-      - name: Login to container registry
-        uses: docker/login-action@v2.1.0
+      - name: Login to GitHub container registry
+        uses: docker/login-action@v2.2.0
         with:
           registry: ghcr.io
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Login to DockerHub container registry
+        uses: docker/[email protected]
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_PASSWORD }}
       - name: Get base registry
         run: |
           echo "REGISTRY=ghcr.io/${GITHUB_REPOSITORY,,}" >> $GITHUB_ENV
@@ -67,7 +72,8 @@ jobs:
         uses: docker/[email protected]
         with:
           context: ${{ inputs.folder }}
-          build-args: ${{ inputs.build-args }}
+          build-args: |-
+            ${{ inputs.build-args }}
           push: true
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
@@ -80,10 +86,10 @@ jobs:
       - name: Comment
         if: steps.PR.outputs.number
         uses: peter-evans/[email protected]
-        with: 
+        with:
           issue-number: ${{ steps.PR.outputs.number }}
           body: >
-            @${{ github.triggering_actor }} Build complete, ${{ steps.docker-build.outcome }}: 
+            @${{ github.triggering_actor }} Build complete, ${{ steps.docker-build.outcome }}:
             ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
 
-            Image: `${{ fromJSON(steps.docker-build.outputs.metadata)['image.name'] }}`
+            Image: `${{ fromJSON(steps.docker-build.outputs.metadata)['image.name'] }}`

.github/workflows/cuda-ssh.yml

@@ -16,6 +16,7 @@ jobs:
           - ceeb8c2-nccl-cuda11.8.0-nccl2.16.2-1-torch2.0.1-vision0.15.2-audio2.0.2
 
     uses: ./.github/workflows/build.yml
+    secrets: inherit
     with:
       image-name: cuda-ssh
       folder: cuda-ssh

.github/workflows/gpt-neox-determined.yml

@@ -10,7 +10,8 @@ on:
 jobs:
   build:
     uses: ./.github/workflows/build.yml
-    with: 
+    secrets: inherit
+    with:
       image-name: gpt-neox-determined
       folder: gpt-neox-determined
-      build-args: ""
+      build-args: ""

.github/workflows/gpt-neox-mpi.yml

@@ -10,6 +10,7 @@ on:
 jobs:
   build:
     uses: ./.github/workflows/build.yml
+    secrets: inherit
     with:
       image-name: gpt-neox-mpi
       folder: gpt-neox-mpi

.github/workflows/read-configuration.yml

@@ -0,0 +1,45 @@
+name: read-configuration
+
+on:
+  workflow_call:
+    inputs:
+      path:
+        required: true
+        type: string
+      filter:
+        required: false
+        type: string
+    outputs:
+      config:
+        description: "The retrieved configuration, as JSON"
+        value: ${{ jobs.read-file.outputs.config }}
+
+jobs:
+  read-file:
+    name: Read Configuration File
+    runs-on: ["self-hosted", "Linux"]
+    permissions: {}
+    outputs:
+      config: ${{ steps.read.outputs.contents }}
+    steps:
+      - uses: actions/checkout@v3
+      - name: Read configuration
+        id: read
+        env:
+          FILE_PATH: ${{ inputs.path }}
+          FILTER: ${{ inputs.filter }}
+        run: |
+          set -x;
+          if [ -n "$FILTER" ]; then
+            CONTENTS="$(yq e "$FILE_PATH" --expression "$FILTER" -oj -I0)";
+          else
+            CONTENTS="$(yq e "$FILE_PATH" -oj -I0)";
+          fi;
+          echo "contents=$CONTENTS" >> "$GITHUB_OUTPUT";
+          
+          {
+            echo '## Configuration';
+            echo '```json';
+            echo "$CONTENTS" | jq .;
+            echo '```';
+          } >> "$GITHUB_STEP_SUMMARY";

.github/workflows/sd-finetuner.yml

@@ -15,7 +15,8 @@ on:
 jobs:
   build:
     uses: ./.github/workflows/build.yml
-    with: 
+    secrets: inherit
+    with:
       image-name: sd-finetuner
       folder: sd-finetuner
-      build-args: "--build-arg COMMIT=${{ github.event.inputs.commit }}"
+      build-args: "--build-arg COMMIT=${{ github.event.inputs.commit }}"

.github/workflows/sd-inference.yml

@@ -15,7 +15,9 @@ on:
 jobs:
   build:
     uses: ./.github/workflows/build.yml
-    with: 
+    secrets: inherit
+    with:
       image-name: sd-inference
       folder: sd-inference
-      build-args: "--build-arg COMMIT=${{ github.event.inputs.commit }}"
+      build-args: |
+        COMMIT=${{ github.event.inputs.commit }}

.github/workflows/slurm.yml

@@ -21,7 +21,8 @@ jobs:
             BASE_IMAGE=registry.gitlab.com/coreweave/sunk/slurmd-cw-cu117-extras:bc5a133d
 
     uses: ./.github/workflows/build.yml
-    with: 
+    secrets: inherit
+    with:
       image-name: ${{ matrix.image.name }}
       folder: ${{ matrix.image.folder }}
       build-args: ${{ matrix.image.build-args }}

.github/workflows/tensorizer.yml

@@ -15,7 +15,8 @@ on:
 jobs:
   build:
     uses: ./.github/workflows/build.yml
-    with: 
+    secrets: inherit
+    with:
       image-name: tensorizer
       folder: tensorizer
-      build-args: "--build-arg COMMIT=${{ github.event.inputs.commit }}"
+      build-args: "--build-arg COMMIT=${{ github.event.inputs.commit }}"

.github/workflows/torch-base.yml

@@ -1,26 +1,41 @@
+name: torch-base
+
 on:
   workflow_dispatch:
+    inputs:
+      image-name:
+        required: false
+        description: "Custom name under which to publish the resulting container"
+        type: string
+      image-tag-suffix:
+        required: false
+        description: "Custom tag suffix listing library versions under which to publish the resulting container"
+        type: string
   push:
     paths:
       - "torch/**"
+      - ".github/configurations/torch-base.yml"
       - ".github/workflows/torch-base.yml"
       - ".github/workflows/torch.yml"
       - ".github/workflows/build.yml"
 
 
 jobs:
+  get-config:
+    name: Get torch:base Config
+    uses: ./.github/workflows/read-configuration.yml
+    with:
+      path: ./.github/configurations/torch-base.yml
   build:
+    name: Build torch:base
+    needs: get-config
     strategy:
-      matrix:
-        cuda: [12.1.1, 12.0.1, 11.8.0]
-        include:
-          - torch: 2.0.1
-            vision: 0.15.2
-            audio: 2.0.2
-
+      matrix: ${{ fromJSON(needs.get-config.outputs.config) }}
     uses: ./.github/workflows/torch.yml
+    secrets: inherit
     with:
-      tag: ${{ format('base-cuda{0}-torch{1}-vision{2}-audio{3}', matrix.cuda, matrix.torch, matrix.vision, matrix.audio) }}
+      image-name: ${{ inputs.image-name }}
+      tag: ${{ format('{0}-{1}', format('base-cuda{0}', matrix.cuda), inputs.image-tag-suffix || format('torch{0}-vision{1}-audio{2}', matrix.torch, matrix.vision, matrix.audio)) }}
       builder-base-image: nvidia/cuda:${{ matrix.cuda }}-devel-ubuntu20.04
       base-image: nvidia/cuda:${{ matrix.cuda }}-base-ubuntu20.04
       torch-version: ${{ matrix.torch }}