feat: add authorization for self-hosted runner

.github/workflows/test-self-hosted-linux.yml

@@ -23,7 +23,7 @@ jobs:
       fail-fast: false
       matrix:
         flash-attn-version: ["2.8.0"]
-        python-version: ["3.11"]
+        python-version: ["3.11", "3.12"]
         torch-version: ["2.7.1"]
         # https://developer.nvidia.com/cuda-toolkit-archive
         cuda-version: ["12.8.1"]
@@ -92,6 +92,7 @@ jobs:
 
       - name: Build wheels
         timeout-minutes: 1200
+        id: build_wheels
         shell: bash
         env:
           MAX_JOBS: 4
@@ -105,7 +106,8 @@ jobs:
       - name: Install Test
         shell: bash
         run: |
-          pip install --no-cache-dir --force-reinstall flash-attention/dist/${{ env.wheel_name }}
+          pip uninstall -y flash-attn > /dev/null 2>&1
+          pip install --no-cache-dir flash-attention/dist/${{ steps.build_wheels.outputs.WHEEL_NAME }}
           python -c "import flash_attn; print(flash_attn.__version__)"
 
       - name: Clean up
@@ -113,4 +115,3 @@ jobs:
         if: always()
         run: |
           rm -rf /opt/hostedtoolcache/Python
-          rm -rf ~/.cache/pip

self-hosted-runner/Dockerfile

@@ -1,12 +1,13 @@
 FROM ubuntu:24.04
 
-ARG REPOSITORY_URL
-ARG PERSONAL_ACCESS_TOKEN
-ARG GH_RUNNER_VERSION="2.324.0"
+ARG REPOSITORY_URL="https://github.com/user/repo"
+ARG GH_RUNNER_VERSION="2.329.0"
 ARG RUNNER_NAME="self-hosted-github-actions-runner"
 ARG RUNNER_GROUP="default"
 ARG RUNNER_LABELS="self-hosted,Linux"
 ARG TARGET_ARCH="x64"
+ARG PERSONAL_ACCESS_TOKEN=""
+ARG REGISTRY_TOKEN=""
 
 ENV DEBIAN_FRONTEND=noninteractive \
     AGENT_TOOLSDIRECTORY=/opt/hostedtoolcache
@@ -65,14 +66,26 @@ RUN curl -fsSL -o actions-runner.tar.gz -L "https://github.com/actions/runner/re
     && tar xf actions-runner.tar.gz \
     && rm actions-runner.tar.gz \
     && sudo ./bin/installdependencies.sh \
-    && ./config.sh \
+    && if [ -n "$PERSONAL_ACCESS_TOKEN" ]; then \
+    ./config.sh \
     --unattended \
     --url $REPOSITORY_URL \
-    --pat $PERSONAL_ACCESS_TOKEN \
+    --pat "$PERSONAL_ACCESS_TOKEN" \
     --name $RUNNER_NAME \
     --runnergroup $RUNNER_GROUP \
     --labels "${RUNNER_LABELS},${TARGET_ARCH}" \
     --work /home/ubuntu/actions-runner \
-    --replace
+    --replace; \
+    else \
+    ./config.sh \
+    --unattended \
+    --url $REPOSITORY_URL \
+    --token "$REGISTRY_TOKEN" \
+    --name $RUNNER_NAME \
+    --runnergroup $RUNNER_GROUP \
+    --labels "${RUNNER_LABELS},${TARGET_ARCH}" \
+    --work /home/ubuntu/actions-runner \
+    --replace; \
+    fi
 
 CMD ["./run.sh"]

self-hosted-runner/compose.yml

@@ -2,14 +2,16 @@ services:
   runner:
     privileged: true
     restart: always
+    env_file:
+      - .env
     build:
       context: .
       dockerfile: Dockerfile
       args:
         REPOSITORY_URL: https://github.com/mjun0812/flash-attention-prebuild-wheels
         PERSONAL_ACCESS_TOKEN: $PERSONAL_ACCESS_TOKEN
-        GH_RUNNER_VERSION: 2.327.1
+        GH_RUNNER_VERSION: 2.329.0
         RUNNER_NAME: self-hosted-runner
         RUNNER_GROUP: default
-        RUNNER_LABELS: self-hosted
+        RUNNER_LABELS: self-hosted,Linux
         TARGET_ARCH: x64

self-hosted-runner/env.template

@@ -1 +1,3 @@
 PERSONAL_ACCESS_TOKEN=
+REGISTRY_TOKEN=
+RUNNER_LABELS=Linux,self-hosted