Merge branch 'dev' of https://github.com/menloresearch/cortex.cpp into s/chore/tool

sangjanai · sangjanai · commit a256c2ac96e4 · 2025-04-11T13:26:17.000+07:00
diff --git a/.github/workflows/template-build-macos.yml b/.github/workflows/template-build-macos.yml
@@ -253,6 +253,14 @@ jobs:
           cd engine
           make codesign-binary CODE_SIGN=true DEVELOPER_ID="${{ secrets.DEVELOPER_ID }}" DESTINATION_BINARY_NAME="${{ steps.set-output-params.outputs.destination_binary_name }}" DESTINATION_BINARY_SERVER_NAME="${{ steps.set-output-params.outputs.destination_binary_server_name }}"
 
+      - name: Code Signing binaries for separate binary
+        run: |
+          codesign --force -s "${{ secrets.DEVELOPER_ID }}" --options=runtime --entitlements="./engine/templates/macos/entitlements.plist" ./cortex-${{ inputs.new_version }}-mac-arm64/${{ steps.set-output-params.outputs.destination_binary_name }}
+          codesign --force -s "${{ secrets.DEVELOPER_ID }}" --options=runtime --entitlements="./engine/templates/macos/entitlements.plist" ./cortex-${{ inputs.new_version }}-mac-arm64/${{ steps.set-output-params.outputs.destination_binary_server_name }}
+
+          codesign --force -s "${{ secrets.DEVELOPER_ID }}" --options=runtime --entitlements="./engine/templates/macos/entitlements.plist" ./cortex-${{ inputs.new_version }}-mac-amd64/${{ steps.set-output-params.outputs.destination_binary_name }}
+          codesign --force -s "${{ secrets.DEVELOPER_ID }}" --options=runtime --entitlements="./engine/templates/macos/entitlements.plist" ./cortex-${{ inputs.new_version }}-mac-amd64/${{ steps.set-output-params.outputs.destination_binary_server_name }}
+
       - name: Notary macOS Binary
         run: |
           curl -sSfL https://raw.githubusercontent.com/anchore/quill/main/install.sh | sh -s -- -b /usr/local/bin
@@ -265,6 +273,18 @@ jobs:
           QUILL_NOTARY_ISSUER: ${{ secrets.NOTARY_ISSUER }}
           QUILL_NOTARY_KEY: "/tmp/notary-key.p8"
 
+      - name: Notary macOS Binary for separate binary
+        run: |
+          # Notarize the binary
+          quill notarize ./cortex-${{ inputs.new_version }}-mac-arm64/${{ steps.set-output-params.outputs.destination_binary_name }}
+          quill notarize ./cortex-${{ inputs.new_version }}-mac-arm64/${{ steps.set-output-params.outputs.destination_binary_server_name }}
+          quill notarize ./cortex-${{ inputs.new_version }}-mac-amd64/${{ steps.set-output-params.outputs.destination_binary_name }}
+          quill notarize ./cortex-${{ inputs.new_version }}-mac-amd64/${{ steps.set-output-params.outputs.destination_binary_server_name }}
+        env:
+          QUILL_NOTARY_KEY_ID: ${{ secrets.NOTARY_KEY_ID }}
+          QUILL_NOTARY_ISSUER: ${{ secrets.NOTARY_ISSUER }}
+          QUILL_NOTARY_KEY: "/tmp/notary-key.p8"
+
       - name: Build network Installers
         shell: bash
         run: |
@@ -310,6 +330,24 @@ jobs:
           xcrun notarytool submit ${{ steps.set-output-params.outputs.package_name }}-local.pkg --apple-id ${{ secrets.APPLE_ID }} --password ${{ secrets.APPLE_APP_SPECIFIC_PASSWORD }} --team-id ${{ secrets.APPLE_TEAM_ID }} --wait
 
       - name: Package
+        run: |
+          mkdir temp
+          # Mac arm64
+          mv cortex-${{ inputs.new_version }}-mac-arm64 temp/cortex
+          cd temp
+          tar -czvf cortex-arm64.tar.gz cortex
+          mv cortex-arm64.tar.gz ../cortex-arm64.tar.gz
+          cd ..
+          rm -rf temp/cortex
+
+          # Mac amd64
+          mv cortex-${{ inputs.new_version }}-mac-amd64 temp/cortex
+          cd temp
+          tar -czvf cortex-amd64.tar.gz cortex
+          mv cortex-amd64.tar.gz ../cortex-amd64.tar.gz
+          cd ..
+
+      - name: Package for separate binary
         run: |
           cd engine
           make package
@@ -320,6 +358,18 @@ jobs:
           name: cortex-${{ inputs.new_version }}-mac-universal
           path: ./engine/cortex
   
+      - name: Upload Artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: cortex-${{ inputs.new_version }}-mac-arm64-signed
+          path: ./cortex-${{ inputs.new_version }}-mac-arm64
+
+      - name: Upload Artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: cortex-${{ inputs.new_version }}-mac-amd64-signed
+          path: ./cortex-${{ inputs.new_version }}-mac-amd64
+
       - name: Upload Artifact
         uses: actions/upload-artifact@v4
         with:
@@ -358,6 +408,28 @@ jobs:
           asset_name: cortex-${{ inputs.new_version }}-mac-universal.tar.gz
           asset_content_type: application/zip
 
+      - name: Upload release assert if public provider is github
+        if: inputs.public_provider == 'github'
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        uses: actions/upload-release-asset@v1.0.1
+        with:
+          upload_url: ${{ inputs.upload_url }}
+          asset_path: ./cortex-arm64.tar.gz
+          asset_name: cortex-${{ inputs.new_version }}-mac-arm64.tar.gz
+          asset_content_type: application/zip
+
+      - name: Upload release assert if public provider is github
+        if: inputs.public_provider == 'github'
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        uses: actions/upload-release-asset@v1.0.1
+        with:
+          upload_url: ${{ inputs.upload_url }}
+          asset_path: ./cortex-amd64.tar.gz
+          asset_name: cortex-${{ inputs.new_version }}-mac-amd64.tar.gz
+          asset_content_type: application/zip
+
       - name: Upload release assert if public provider is github
         if: inputs.public_provider == 'github'
         env:
diff --git a/engine/extensions/local-engine/local_engine.cc b/engine/extensions/local-engine/local_engine.cc
@@ -20,7 +20,7 @@ const std::unordered_set<std::string> kIgnoredParams = {
     "user_prompt",  "min_keep",        "mirostat",   "mirostat_eta",
     "mirostat_tau", "text_model",      "version",    "n_probs",
     "object",       "penalize_nl",     "precision",  "size",
-    "stop",         "tfs_z",           "typ_p"};
+    "stop",         "tfs_z",           "typ_p",      "caching_enabled"};
 
 const std::unordered_map<std::string, std::string> kParamsMap = {
     {"cpu_threads", "--threads"},
@@ -67,6 +67,19 @@ std::vector<std::string> ConvertJsonToParamsVector(const Json::Value& root) {
         res.push_back("--embedding");
       }
       continue;
+    } else if (member == "cache_type") {
+      if (!root[member].isNull()) {
+        res.push_back("-ctk");
+        res.push_back(root[member].asString());
+        res.push_back("-ctv");
+        res.push_back(root[member].asString());
+      }
+      continue;
+    } else if (member == "use_mmap") {
+      if (!root[member].asBool()) {
+        res.push_back("--no-mmap");
+      }
+      continue;
     }
 
     res.push_back("--" + member);
diff --git a/engine/services/engine_service.cc b/engine/services/engine_service.cc
@@ -772,7 +772,13 @@ EngineService::GetInstalledEngineVariants(const std::string& engine) const {
         // try to find version.txt
         auto version_txt_path = version_entry.path() / "version.txt";
         if (!std::filesystem::exists(version_txt_path)) {
-          continue;
+          // create new one
+          std::ofstream meta(version_txt_path, std::ios::out);
+          meta << "name: " << entry.path().filename() << std::endl;
+          meta << "version: " << version_entry.path().filename() << std::endl;
+          meta.close();
+          CTL_INF("name: " << entry.path().filename().string() << ", version: "
+                           << version_entry.path().filename().string());
         }
 
         try {
@@ -865,7 +871,9 @@ void EngineService::RegisterEngineLibPath() {
 
       // register deps
       std::vector<std::filesystem::path> paths{};
-      paths.push_back(cuda_path);
+      if (std::filesystem::exists(cuda_path)) {
+        paths.push_back(cuda_path);
+      }
       paths.push_back(engine_dir_path);
 
       CTL_DBG("Registering dylib for "
