Merge main into fix/whispercpp-vad

.gitattributes

@@ -0,0 +1 @@
+* text=auto eol=lf

.github/workflows/on-merge-qvac-lib-infer-llamacpp-embed.yml

@@ -84,6 +84,7 @@ jobs:
     with:
       tag: ${{ inputs.tag || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-llamacpp-embed"
 
   publish-feature-gpr:
     needs: publish-logic
@@ -97,6 +98,7 @@ jobs:
     with:
       tag: ${{ inputs.tag || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-llamacpp-embed"
 
   publish-tmp-gpr:
     needs: publish-logic
@@ -110,6 +112,7 @@ jobs:
     with:
       tag: ${{ inputs.tag || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-llamacpp-embed"
 
   publish-release-npm:
     needs: publish-logic

.github/workflows/on-merge-qvac-lib-infer-llamacpp-llm.yml

@@ -92,6 +92,7 @@ jobs:
     with:
       tag: ${{ inputs.tag || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-llamacpp-llm"
 
   publish-feature-gpr:
     needs: publish-logic
@@ -105,6 +106,7 @@ jobs:
     with:
       tag: ${{ inputs.tag || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-llamacpp-llm"
 
   publish-tmp-gpr:
     needs: publish-logic
@@ -118,6 +120,7 @@ jobs:
     with:
       tag: ${{ inputs.tag || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-llamacpp-llm"
 
   publish-release-npm:
     needs: publish-logic

.github/workflows/on-merge-qvac-lib-infer-whispercpp.yml

@@ -91,6 +91,7 @@ jobs:
       # workflow_dispatch -> github.event.inputs.tag is set; push -> fallback to computed gpr_tag
       tag: ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.tag) || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-whispercpp"
 
   publish-feature-gpr:
     needs: publish-logic
@@ -104,6 +105,7 @@ jobs:
     with:
       tag: ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.tag) || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-whispercpp"
 
   publish-tmp-gpr:
     needs: publish-logic
@@ -117,6 +119,7 @@ jobs:
     with:
       tag: ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.tag) || needs.publish-logic.outputs.gpr_tag }}
       publish_target: "gpr"
+      workdir: "packages/qvac-lib-infer-whispercpp"
 
   publish-release-npm:
     needs: publish-logic

.github/workflows/prebuilds-qvac-lib-infer-llamacpp-embed.yml

@@ -468,7 +468,7 @@ jobs:
         uses: actions/download-artifact@v4
         with:
           name: prebuilds
-          path: prebuilds
+          path: ${{ env.WORKDIR }}/prebuilds
 
       - name: Publish to GitHub Package Registry
         id: publish
@@ -510,11 +510,12 @@ jobs:
         uses: actions/download-artifact@v4
         with:
           name: prebuilds
-          path: prebuilds
+          path: ${{ env.WORKDIR }}/prebuilds
 
       - name: Publish to NPM Package Registry
         id: publish
         uses: tetherto/qvac-devops/.github/actions/publish-library-to-npm@monorepo_update
         with:
           secret-token: ${{ secrets.NPM_TOKEN }}
           tag: "latest"
+          workdir: ${{ env.WORKDIR }}

.github/workflows/prebuilds-qvac-lib-infer-llamacpp-llm.yml

@@ -495,7 +495,7 @@ jobs:
         uses: actions/download-artifact@v4
         with:
           name: prebuilds
-          path: prebuilds
+          path: ${{ env.WORKDIR }}/prebuilds
 
       - name: Scope package as @tetherto for GPR
         shell: bash
@@ -549,10 +549,11 @@ jobs:
         uses: actions/download-artifact@v4
         with:
           name: prebuilds
-          path: prebuilds
+          path: ${{ env.WORKDIR }}/prebuilds
 
       # Re-assert scope for public npm (fresh checkout)
       - name: Scope package as @qvac for npm
+        working-directory: ${{ env.WORKDIR }}
         shell: bash
         run: |
           set -euo pipefail
@@ -564,4 +565,5 @@ jobs:
         uses: tetherto/qvac-devops/.github/actions/publish-library-to-npm@monorepo_update
         with:
           secret-token: ${{ secrets.NPM_TOKEN }}
-          tag: 'latest'
+          tag: 'latest'
+          workdir: ${{ env.WORKDIR }}

.github/workflows/prebuilds-qvac-lib-infer-nmtcpp.yml

@@ -292,23 +292,41 @@ jobs:
           sudo apt install -y vulkan-sdk
 
       - if: ${{ matrix.os == 'ubuntu-24.04-arm64-private' }}
-        name: Cache Vulkan SDK for linux arm64
-        id: cache-vulkan-arm64
-        uses: actions/cache@v4
-        with:
-          path: ~/vulkan
-          key: vulkan-sdk-linux-arm64-v1
-
-      - if: ${{ matrix.os == 'ubuntu-24.04-arm64-private' && steps.cache-vulkan-arm64.outputs.cache-hit != 'true' }}
-        name: Build Vulkan SDK for linux arm64
+        name: Build Vulkan SDK for linux arm64 (with S3 cache)
+        env:
+          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
+          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          AWS_DEFAULT_REGION: us-east-1
         run: |
           sudo apt install -y xz-utils
           wget -q -O /tmp/vulkansdk.tar.xz https://sdk.lunarg.com/sdk/download/latest/linux/vulkan_sdk.tar.xz
           cd ~
           mkdir -p vulkan
           cd vulkan
           tar xf /tmp/vulkansdk.tar.xz --strip-components=1
-          ./vulkansdk --maxjobs
+
+          # Extract SDK major.minor version from README.txt (e.g., "1.4" from "1.4.341.0")
+          SDK_VERSION=$(grep -o 'sdk/[0-9]*\.[0-9]*' README.txt | head -1 | sed 's|sdk/||')
+          S3_BUCKET="tether-ai-dev"
+          S3_KEY="vulkan-sdk-cache/linux-arm64-${SDK_VERSION}.tar.gz"
+
+          echo "Vulkan SDK version: ${SDK_VERSION}"
+
+          # Try to download cached build from S3
+          if aws s3 cp "s3://${S3_BUCKET}/${S3_KEY}" /tmp/vulkan-arm64-cache.tar.gz 2>/dev/null; then
+            echo "Found cached Vulkan SDK, extracting..."
+            tar xzf /tmp/vulkan-arm64-cache.tar.gz -C ~/vulkan
+            rm /tmp/vulkan-arm64-cache.tar.gz
+          else
+            echo "No cache found, building Vulkan SDK for ARM64..."
+            ./vulkansdk --maxjobs
+
+            # Upload the compiled SDK to S3 for future runs
+            echo "Uploading compiled SDK to S3..."
+            tar czf /tmp/vulkan-arm64-cache.tar.gz aarch64
+            aws s3 cp /tmp/vulkan-arm64-cache.tar.gz "s3://${S3_BUCKET}/${S3_KEY}"
+            rm /tmp/vulkan-arm64-cache.tar.gz
+          fi
 
       - if: ${{ matrix.os == 'ubuntu-24.04-arm64-private' }}
         name: Setup Vulkan SDK environment for linux arm64

.github/workflows/prebuilds-qvac-lib-infer-whispercpp.yml

@@ -323,11 +323,38 @@ jobs:
 
       - if: ${{ matrix.platform == 'linux' && matrix.arch == 'arm64' }}
         name: Setup vulkan sdk path in linux arm64
+        env:
+          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
+          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          AWS_DEFAULT_REGION: us-east-1
         run: |
           VULKAN_SDK=~/vulkan/aarch64
           echo "VULKAN_SDK=$VULKAN_SDK" >> $GITHUB_ENV
+
           cd ~/vulkan
-          ./vulkansdk --maxjobs
+
+          # Extract SDK major.minor version from README.txt (e.g., "1.4" from "1.4.341.0")
+          SDK_VERSION=$(grep -o 'sdk/[0-9]*\.[0-9]*' README.txt | head -1 | sed 's|sdk/||')
+          S3_BUCKET="tether-ai-dev"
+          S3_KEY="vulkan-sdk-cache/linux-arm64-${SDK_VERSION}.tar.gz"
+
+          echo "Vulkan SDK version: ${SDK_VERSION}"
+
+          # Try to download cached build from S3
+          if aws s3 cp "s3://${S3_BUCKET}/${S3_KEY}" /tmp/vulkan-arm64-cache.tar.gz 2>/dev/null; then
+            echo "Found cached Vulkan SDK, extracting..."
+            tar xzf /tmp/vulkan-arm64-cache.tar.gz -C ~/vulkan
+            rm /tmp/vulkan-arm64-cache.tar.gz
+          else
+            echo "No cache found, building Vulkan SDK for ARM64..."
+            ./vulkansdk --maxjobs
+
+            # Upload the compiled SDK to S3 for future runs
+            echo "Uploading compiled SDK to S3..."
+            tar czf /tmp/vulkan-arm64-cache.tar.gz aarch64
+            aws s3 cp /tmp/vulkan-arm64-cache.tar.gz "s3://${S3_BUCKET}/${S3_KEY}"
+            rm /tmp/vulkan-arm64-cache.tar.gz
+          fi
 
       - if: ${{ matrix.platform == 'linux' }}
         name: Export rest of vulkan sdk variables in linux
@@ -462,15 +489,7 @@ jobs:
         uses: actions/download-artifact@v4
         with:
           name: prebuilds
-          path: /tmp/prebuilds
-
-      - name: Place merged prebuilds into package workdir
-        shell: bash
-        run: |
-          set -euo pipefail
-          mkdir -p "${WORKDIR}/prebuilds"
-          rm -rf "${WORKDIR}/prebuilds"/*
-          cp -R /tmp/prebuilds/* "${WORKDIR}/prebuilds/"
+          path: ${WORKDIR}/prebuilds
 
       - name: Rename prebuild files from qvac_ to tetherto_ prefix
         shell: bash
@@ -545,12 +564,19 @@ jobs:
           token: ${{ secrets.PAT_TOKEN }}
           fetch-depth: 0
 
+      - name: Download prebuild artifacts
+        uses: actions/download-artifact@v4
+        with:
+          name: prebuilds
+          path: ${WORKDIR}/prebuilds
+
       - name: Publish to NPM Package Registry
         id: publish
         uses: tetherto/qvac-devops/.github/actions/publish-library-to-npm@monorepo_update
         with:
           secret-token: ${{ secrets.NPM_TOKEN }}
           tag: "latest"
+          workdir: $((WORKDIR))
 
       - name: Capture published version
         id: capture_version