diff --git a/.github/workflows/_linux-build.yml b/.github/workflows/_linux-build.yml
index 36a78014087..bed3ce93407 100644
--- a/.github/workflows/_linux-build.yml
+++ b/.github/workflows/_linux-build.yml
@@ -39,7 +39,7 @@ on:
         type: string
         default: "linux.2xlarge"
         description: |
-          List of CUDA architectures CI build should target.
+          Label of the runner this job should run on.
       test-matrix:
         required: false
         type: string
diff --git a/.github/workflows/_win-build.yml b/.github/workflows/_win-build.yml
index 72112c99e7d..6ee529f38b5 100644
--- a/.github/workflows/_win-build.yml
+++ b/.github/workflows/_win-build.yml
@@ -30,6 +30,12 @@ on:
           An option JSON description of what test configs to run later on. This
           is moved here from the Linux test workflow so that we can apply filter
           logic using test-config labels earlier and skip unnecessary builds
+      runner:
+        required: false
+        type: string
+        default: "windows.4xlarge.nonephemeral"
+        description: |
+          Label of the runner this job should run on.
 
     outputs:
       test-matrix:
@@ -43,7 +49,7 @@ jobs:
   build:
     # Don't run on forked repos.
     if: github.repository_owner == 'pytorch'
-    runs-on: [self-hosted, windows.4xlarge.nonephemeral]
+    runs-on: ${{ inputs.runner }}
     timeout-minutes: 240
     outputs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
diff --git a/.github/workflows/pull.yml b/.github/workflows/pull.yml
index 472bb4fa5b6..c0e75d1f503 100644
--- a/.github/workflows/pull.yml
+++ b/.github/workflows/pull.yml
@@ -383,15 +383,17 @@ jobs:
     if: github.event_name == 'pull_request'
     name: win-vs2019-cpu-py3
     uses: ./.github/workflows/_win-build.yml
+    needs: get-label-type
     with:
       build-environment: win-vs2019-cpu-py3
       cuda-version: cpu
       sync-tag: win-cpu-build
+      runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 3, runner: "windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 2, num_shards: 3, runner: "windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 3, num_shards: 3, runner: "windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
         ]}
 
   linux-focal-cpu-py3_10-gcc9-bazel-test:
diff --git a/.github/workflows/trunk.yml b/.github/workflows/trunk.yml
index 946aec33445..5ef246c22f8 100644
--- a/.github/workflows/trunk.yml
+++ b/.github/workflows/trunk.yml
@@ -179,15 +179,17 @@ jobs:
   win-vs2019-cpu-py3-build:
     name: win-vs2019-cpu-py3
     uses: ./.github/workflows/_win-build.yml
+    needs: get-label-type
     with:
       build-environment: win-vs2019-cpu-py3
       cuda-version: cpu
       sync-tag: win-cpu-build
+      runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 3, runner: "windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 2, num_shards: 3, runner: "windows.4xlarge.nonephemeral" },
-          { config: "default", shard: 3, num_shards: 3, runner: "windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 1, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 2, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 3, num_shards: 3, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
         ]}
 
   win-vs2019-cpu-py3-test:
@@ -204,19 +206,21 @@ jobs:
   win-vs2019-cuda11_8-py3-build:
     name: win-vs2019-cuda11.8-py3
     uses: ./.github/workflows/_win-build.yml
+    needs: get-label-type
     with:
       build-environment: win-vs2019-cuda11.8-py3
       cuda-version: "11.8"
       sync-tag: win-cuda-build
+      runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 6, runner: "windows.g5.4xlarge.nvidia.gpu" },
-          { config: "default", shard: 2, num_shards: 6, runner: "windows.g5.4xlarge.nvidia.gpu" },
-          { config: "default", shard: 3, num_shards: 6, runner: "windows.g5.4xlarge.nvidia.gpu" },
-          { config: "default", shard: 4, num_shards: 6, runner: "windows.g5.4xlarge.nvidia.gpu" },
-          { config: "default", shard: 5, num_shards: 6, runner: "windows.g5.4xlarge.nvidia.gpu" },
-          { config: "default", shard: 6, num_shards: 6, runner: "windows.g5.4xlarge.nvidia.gpu" },
-          { config: "force_on_cpu", shard: 1, num_shards: 1, runner: "windows.4xlarge.nonephemeral" },
+          { config: "default", shard: 1, num_shards: 6, runner: "${{ needs.get-label-type.outputs.label-type }}windows.g5.4xlarge.nvidia.gpu" },
+          { config: "default", shard: 2, num_shards: 6, runner: "${{ needs.get-label-type.outputs.label-type }}windows.g5.4xlarge.nvidia.gpu" },
+          { config: "default", shard: 3, num_shards: 6, runner: "${{ needs.get-label-type.outputs.label-type }}windows.g5.4xlarge.nvidia.gpu" },
+          { config: "default", shard: 4, num_shards: 6, runner: "${{ needs.get-label-type.outputs.label-type }}windows.g5.4xlarge.nvidia.gpu" },
+          { config: "default", shard: 5, num_shards: 6, runner: "${{ needs.get-label-type.outputs.label-type }}windows.g5.4xlarge.nvidia.gpu" },
+          { config: "default", shard: 6, num_shards: 6, runner: "${{ needs.get-label-type.outputs.label-type }}windows.g5.4xlarge.nvidia.gpu" },
+          { config: "force_on_cpu", shard: 1, num_shards: 1, runner: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral" },
         ]}
 
   linux-focal-rocm6_1-py3_8-build: