(vllm) Version fixes

.github/actions/build-vllm/action.yml

@@ -23,6 +23,27 @@ runs:
     - name: Expose cache environment variables
       uses: ./.github/actions/common/expose-cache-vars
 
+    - name: Compute vLLM version
+      uses: actions/github-script@v7
+      with:
+        script: |
+          // ref
+          const ref = core.getInput('ref');
+
+          // export version for setuptools_scm
+          if (ref.startsWith("v")) {
+            core.exportVariable('SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM', ref.slice(1));
+          } else if (ref === "main") {
+            core.exportVariable('SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM', '999.999.999');
+          }
+
+          // export version for docker
+          if (ref === "main") {
+            core.exportVariable('DOCKER_TAGS', `ghcr.io/${context.repo.owner}/vllm:latest`);
+          } else {
+            core.exportVariable('DOCKER_TAGS', `ghcr.io/${context.repo.owner}/vllm:${ref}`)
+          }
+
     - name: Checkout ${{ inputs.repository }} (${{ inputs.ref }})
       uses: actions/checkout@v4
       with:
@@ -56,7 +77,7 @@ runs:
         context: ${{ inputs.repository }}/${{ inputs.ref }}
         outputs: type=tar,dest=build.tar
         push: false
-        tags: ghcr.io/${{ github.repository_owner }}/vllm:${{ inputs.ref }}
+        tags: ${{ env.DOCKER_TAGS }}
         target: build
 
         build-args: |
@@ -68,6 +89,7 @@ runs:
         secrets: |
           ACTIONS_CACHE_URL=${{ env.ACTIONS_CACHE_URL }}
           ACTIONS_RUNTIME_TOKEN=${{ env.ACTIONS_RUNTIME_TOKEN }}
+          SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM=${{ env.SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM }}
 
     - name: Copy wheel files
       shell: bash
@@ -94,7 +116,7 @@ runs:
         cache-from: type=local,src=/var/tmp/buildx-cache
         context: ${{ inputs.repository }}/${{ inputs.ref }}
         push: true
-        tags: ghcr.io/${{ github.repository_owner }}/vllm:${{ inputs.ref }}
+        tags: ${{ env.DOCKER_TAGS }}
         target: vllm-openai
 
         build-args: |

.github/workflows/vllm-pascal-main.yml

@@ -28,7 +28,7 @@ jobs:
       - name: Publish wheels
         uses: ./.github/actions/common/publish-wheels
         with:
-          prefix: vllm_pascal-0.1.dev
+          prefix: vllm_pascal-999.999.999+
 
   update-index:
     needs: publish

README.md

@@ -15,8 +15,8 @@ The main repository for building Pascal-compatible versions of ML applications a
 
 ```sh
 # Pull the vLLM image
-docker pull ghcr.io/sasha0552/vllm:v0.6.3  # note that the tag "latest" is not supported.
-                                           # use the tag "main" for the nightly version.
+docker pull ghcr.io/sasha0552/vllm:v0.6.3  # you can omit the version specifier
+                                           # to install nightly version
 
 # You can now follow the official vLLM documentation.
 # Replace the official image with this one.
@@ -72,7 +72,8 @@ python -m venv venv
 source venv/bin/activate
 
 # Install vLLM
-pip3 install vllm-pascal
+pip3 install vllm-pascal==0.6.3  # you can omit the version specifier
+                                 # to install nightly version
 
 # Install patched triton
 transient-package install       \

patches/vllm-project/vllm/v0.5.5/0000-sccache.patch → patches/vllm-project/vllm/v0.5.5/0000-environment-variables.patch

@@ -1,11 +1,12 @@
 --- a/Dockerfile
 +++ b/Dockerfile
-@@ -89,14 +89,15 @@ ARG SCCACHE_BUCKET_NAME=vllm-build-sccache
+@@ -89,14 +89,16 @@ ARG SCCACHE_BUCKET_NAME=vllm-build-sccache
  ARG SCCACHE_REGION_NAME=us-west-2
  # if USE_SCCACHE is set, use sccache to speed up compilation
  RUN --mount=type=cache,target=/root/.cache/pip \
 +    --mount=type=secret,id=ACTIONS_CACHE_URL,env=ACTIONS_CACHE_URL \
 +    --mount=type=secret,id=ACTIONS_RUNTIME_TOKEN,env=ACTIONS_RUNTIME_TOKEN \
++    --mount=type=secret,id=SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM,env=SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM \
      if [ "$USE_SCCACHE" = "1" ]; then \
          echo "Installing sccache..." \
          && curl -L -o sccache.tar.gz https://github.com/mozilla/sccache/releases/download/v0.8.1/sccache-v0.8.1-x86_64-unknown-linux-musl.tar.gz \

patches/vllm-project/vllm/v0.6.2/0000-sccache.patch → patches/vllm-project/vllm/v0.6.2/0000-environment-variables.patch

@@ -1,11 +1,12 @@
 --- a/Dockerfile
 +++ b/Dockerfile
-@@ -86,15 +86,15 @@ ARG SCCACHE_S3_NO_CREDENTIALS=0
+@@ -86,15 +86,16 @@ ARG SCCACHE_S3_NO_CREDENTIALS=0
  # if USE_SCCACHE is set, use sccache to speed up compilation
  RUN --mount=type=cache,target=/root/.cache/pip \
      --mount=type=bind,source=.git,target=.git \
 +    --mount=type=secret,id=ACTIONS_CACHE_URL,env=ACTIONS_CACHE_URL \
 +    --mount=type=secret,id=ACTIONS_RUNTIME_TOKEN,env=ACTIONS_RUNTIME_TOKEN \
++    --mount=type=secret,id=SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM,env=SETUPTOOLS_SCM_PRETEND_VERSION_FOR_VLLM \
      if [ "$USE_SCCACHE" = "1" ]; then \
          echo "Installing sccache..." \
          && curl -L -o sccache.tar.gz https://github.com/mozilla/sccache/releases/download/v0.8.1/sccache-v0.8.1-x86_64-unknown-linux-musl.tar.gz \