cuda 12.1 build (for newer gpus) (#4)

* cuda 12.1 build (for newer gpus) * seperate jobs * seperate jobs * cuda 12.1 use develepment (need nvcc for deepspeed) * remove cache for build space
coqui-ai · Oct 30, 2023 · de53a25 · de53a25
1 parent 634e257
commit de53a25
Show file tree

Hide file tree

Showing 2 changed files with 96 additions and 4 deletions.
diff --git a/.github/workflows/build-and-push-to-ghcr.yml b/.github/workflows/build-and-push-to-ghcr.yml
@@ -4,7 +4,7 @@ on:
     branches: [main]
   pull_request:
 jobs:
-  build-and-push-to-ghcr:
+  build-and-push-to-ghcr-cuda117:
     runs-on: ubuntu-22.04
     steps:
       -
@@ -20,17 +20,24 @@ jobs:
           set -xe
           docker login --username ${{ github.actor }} --password ${{ secrets.GITHUB_TOKEN }} ghcr.io
 
-      - name: Build only for PR
+      - name: 'Remove cache'
+        run: | 
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf "/usr/local/share/boost"
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+
+      - name: Build only for PR Cuda 11.7
         if: github.ref != 'refs/heads/main'
         uses: docker/build-push-action@v5
         with:
           context: "{{defaultContext}}:server"
           file: Dockerfile
           push: false # Do not push image for PR
-          cache-from: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest;type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-${{ github.event.number }}
+          cache-from: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest; type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-${{ github.event.number }}
           cache-to: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-${{ github.event.number }}
 
-      - name: Build and Push image
+      - name: Build and Push image Cuda 11.7
         if: github.ref == 'refs/heads/main'
         uses: docker/build-push-action@v5
         with:
@@ -42,13 +49,39 @@ jobs:
           tags: ghcr.io/coqui-ai/xtts-streaming-server:latest, ghcr.io/coqui-ai/xtts-streaming-server:main-${{ github.sha }}
           #build-args:
 
+  build-and-push-to-ghcr-cuda118:
+    runs-on: ubuntu-22.04
+    steps:
+      -
+        name: Checkout
+        uses: actions/checkout@v3
+
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: 'Login to GitHub Container Registry'
+        run: |
+          set -xe
+          docker login --username ${{ github.actor }} --password ${{ secrets.GITHUB_TOKEN }} ghcr.io
+
       - name: 'Remove cache'
         run: | 
           sudo rm -rf /usr/share/dotnet
           sudo rm -rf /opt/ghc
           sudo rm -rf "/usr/local/share/boost"
           sudo rm -rf "$AGENT_TOOLSDIRECTORY"
 
+      - name: Build only for PR cuda 11.8
+        if: github.ref != 'refs/heads/main'
+        uses: docker/build-push-action@v5
+        with:
+          context: "{{defaultContext}}:server"
+          file: Dockerfile.cuda118
+          push: false # Do not push image for PR
+          cache-from: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest-cuda118; type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-cuda118-${{ github.event.number }}
+          cache-to: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-cuda118-${{ github.event.number }}
+
       - name: Build and Push image cuda 11.8
         if: github.ref == 'refs/heads/main'
         uses: docker/build-push-action@v5
@@ -60,3 +93,48 @@ jobs:
           cache-to: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest-cuda118
           tags: ghcr.io/coqui-ai/xtts-streaming-server:latest-cuda118, ghcr.io/coqui-ai/xtts-streaming-server:main-cuda118-${{ github.sha }}
           #build-args:
+
+  build-and-push-to-ghcr-cuda121:
+    runs-on: ubuntu-22.04
+    steps:
+      -
+        name: Checkout
+        uses: actions/checkout@v3
+
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: 'Login to GitHub Container Registry'
+        run: |
+          set -xe
+          docker login --username ${{ github.actor }} --password ${{ secrets.GITHUB_TOKEN }} ghcr.io
+
+      - name: 'Remove cache'
+        run: | 
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf "/usr/local/share/boost"
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+
+      - name: Build only for PR cuda 12.1
+        if: github.ref != 'refs/heads/main'
+        uses: docker/build-push-action@v5
+        with:
+          context: "{{defaultContext}}:server"
+          file: Dockerfile.cuda121
+          push: false # Do not push image for PR
+          cache-from: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest-cuda121; type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-cuda121-${{ github.event.number }}
+          cache-to: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-pr-cuda121-${{ github.event.number }}
+
+      - name: Build and Push image cuda 12.1
+        if: github.ref == 'refs/heads/main'
+        uses: docker/build-push-action@v5
+        with:
+          context: "{{defaultContext}}:server"
+          file: Dockerfile.cuda121
+          push: true # Push if merged
+          cache-from: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest-cuda121
+          cache-to: type=registry,ref=ghcr.io/coqui-ai/xtts-streaming-server:cache-latest-cuda121
+          tags: ghcr.io/coqui-ai/xtts-streaming-server:latest-cuda121, ghcr.io/coqui-ai/xtts-streaming-server:main-cuda121-${{ github.sha }}
+          #build-args:
diff --git a/server/Dockerfile.cuda121 b/server/Dockerfile.cuda121
@@ -0,0 +1,14 @@
+FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-devel
+WORKDIR /app
+COPY requirements.txt .
+RUN python -m pip install --use-deprecated=legacy-resolver -r requirements.txt \
+    && python -m pip cache purge
+
+COPY main.py .
+
+#Mark this 1 if you have older card
+ENV NVIDIA_DISABLE_REQUIRE=0
+
+ENV NUM_THREADS=2
+EXPOSE 80
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "80"]