diff --git a/.github/build.bat b/.github/build.bat
index d48f0957..a9df77f5 100755
--- a/.github/build.bat
+++ b/.github/build.bat
@@ -6,6 +6,11 @@ REM SPDX-License-Identifier: MIT
 @echo off
 setlocal enabledelayedexpansion
 
+REM The core project (CMakeLists.txt + src\) lives in the `llama\` module of the Maven
+REM reactor. Re-root here once (relative to this script's own location) so cmake
+REM configures the module regardless of the caller's CWD.
+cd /d "%~dp0..\llama" || exit /b 1
+
 REM ---------------------------------------------------------------------------
 REM Optional shared compiler cache: sccache fronting Depot Cache (WebDAV).
 REM Mirrors build.sh's sccache_can_wrap_compiler() probe. Because sccache *is*
diff --git a/.github/build.sh b/.github/build.sh
index 480e2009..7a47ab65 100755
--- a/.github/build.sh
+++ b/.github/build.sh
@@ -5,6 +5,14 @@
 #
 # SPDX-License-Identifier: MIT
 
+# The core project (CMakeLists.txt + src/) lives in the `llama/` module of the Maven
+# reactor. Re-root here once — every native build delegates to this script (incl. the
+# build_cuda_linux.sh / build_opencl_*.sh wrappers that `exec .github/build.sh`) — so
+# cmake configures the module regardless of the caller's CWD. Anchored to this script's
+# own location (via BASH_SOURCE), so it works from the repo root and inside the dockcross
+# container whose workdir is the mounted repo root.
+cd "$(dirname "${BASH_SOURCE[0]}")/../llama" || exit 1
+
 mkdir -p build
 
 # Build parallelism. Defaults to all cores; RAM-limited CI runners (notably GitHub's
diff --git a/.github/workflows/clang-format.yml b/.github/workflows/clang-format.yml
index fee925c6..623b6db5 100644
--- a/.github/workflows/clang-format.yml
+++ b/.github/workflows/clang-format.yml
@@ -31,8 +31,9 @@ jobs:
       - name: Check C++ formatting
         run: |
           clang-format --version
-          # All hand-written C++ sources; the generated JNI header (src/main/cpp/jllama.h,
-          # produced by `javac -h`) is intentionally excluded.
-          files=$(find src/main/cpp src/test/cpp -type f \( -name '*.cpp' -o -name '*.hpp' \) | sort)
+          # All hand-written C++ sources; the generated JNI header (llama/src/main/cpp/jllama.h,
+          # produced by `javac -h`) is intentionally excluded. Sources live in the `llama/`
+          # reactor module; clang-format finds llama/.clang-format by walking up from each file.
+          files=$(find llama/src/main/cpp llama/src/test/cpp -type f \( -name '*.cpp' -o -name '*.hpp' \) | sort)
           echo "Checking:"; echo "$files"
           clang-format --dry-run --Werror $files
diff --git a/.github/workflows/publish.yml b/.github/workflows/publish.yml
index b4ec2ad5..0d4eb6ff 100644
--- a/.github/workflows/publish.yml
+++ b/.github/workflows/publish.yml
@@ -50,7 +50,7 @@ env:
   # at src/test/resources/images/test-image.jpg (see the README in that
   # directory for licensing). No download step is needed; CI just points
   # mvn test at the committed path.
-  VISION_IMAGE_PATH: "src/test/resources/images/test-image.jpg"
+  VISION_IMAGE_PATH: "llama/src/test/resources/images/test-image.jpg"
 permissions:
   contents: read
 jobs:
@@ -133,25 +133,26 @@ jobs:
           java-version: '21'
           distribution: temurin
       - name: Spotless check (fail fast on format violations)
-        run: mvn -B --no-transfer-progress spotless:check
+        run: mvn -B --no-transfer-progress -f llama/pom.xml spotless:check
       - name: SpotBugs check (fail fast on static-analysis findings)
-        run: mvn -B --no-transfer-progress -DskipTests -Denforcer.skip=true compile spotbugs:check
+        run: mvn -B --no-transfer-progress -f llama/pom.xml -DskipTests -Denforcer.skip=true compile spotbugs:check
       - name: Print internal package dependency graph (jdeps, informational)
         continue-on-error: true
         run: |
-          mvn -B --no-transfer-progress -DskipTests -Denforcer.skip=true compile
+          mvn -B --no-transfer-progress -f llama/pom.xml -DskipTests -Denforcer.skip=true compile
           echo "=== internal package dependency graph (jdeps, bytecode) ==="
-          jdeps -verbose:package target/classes | grep 'net.ladenthin.llama' || true
+          jdeps -verbose:package llama/target/classes | grep 'net.ladenthin.llama' || true
 
   # ---------------------------------------------------------------------------
   # Sibling module `llama-langchain4j` (LangChain4j adapters). Pure Java, no native
   # code and no per-classifier matrix: it compiles against the core's stable Java API
   # (identical across every classifier) and the backend is a runtime choice for the
-  # consumer. This job installs the core Java jar, guards that the module version is in
-  # lockstep with the core, then builds + tests the module (Java 17; langchain4j 1.x
-  # baseline). It runs its mapping unit tests; the model-backed integration test
-  # self-skips without a GGUF. `verify` also builds the javadoc/sources jars so a
-  # release-time javadoc break is caught here in PR CI.
+  # consumer. This job installs the parent + core into the local repo, then builds + tests
+  # the module (Java 17; langchain4j 1.x baseline). It runs its mapping unit tests; the
+  # model-backed integration test self-skips without a GGUF. `verify` also builds the
+  # javadoc/sources jars so a release-time javadoc break is caught here in PR CI. Version
+  # lockstep is now guaranteed by construction (both modules inherit the parent's version),
+  # so the old lockstep guard is gone.
   # ---------------------------------------------------------------------------
 
   test-java-llama-langchain4j:
@@ -164,19 +165,9 @@ jobs:
         with:
           java-version: ${{ env.JAVA_VERSION }}
           distribution: temurin
-      - name: Version lockstep guard (module version must equal core version)
-        shell: bash
-        run: |
-          CORE=$(mvn -q -DforceStdout help:evaluate -Dexpression=project.version | tail -n1)
-          MOD=$(mvn -q -DforceStdout -f llama-langchain4j/pom.xml help:evaluate -Dexpression=project.version | tail -n1)
-          echo "core=$CORE module=$MOD"
-          if [ "$CORE" != "$MOD" ]; then
-            echo "::error::Version drift: core ($CORE) != llama-langchain4j ($MOD). Keep llama-langchain4j/pom.xml <version> in lockstep with the root pom.xml."
-            exit 1
-          fi
-      - name: Install core net.ladenthin:llama jar (Java only)
+      - name: Install parent + core net.ladenthin:llama into the local repo (Java only)
         run: >
-          mvn -B --no-transfer-progress -DskipTests -Denforcer.skip=true
+          mvn -B --no-transfer-progress -pl llama -am -DskipTests -Denforcer.skip=true
           -Dspotless.check.skip=true -Dspotbugs.skip=true
           -Dmaven.javadoc.skip=true -Dmaven.source.skip=true -Dgpg.skip=true install
       - name: Build and test llama-langchain4j
@@ -202,7 +193,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: Linux-x86_64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       - name: Restore shared GGUF model cache (populated by download-models; no re-download)
         uses: actions/cache@v6
         with:
@@ -212,9 +203,9 @@ jobs:
         with:
           distribution: 'temurin'
           java-version: ${{ env.JAVA_VERSION }}
-      - name: Install core net.ladenthin:llama jar (bundles the downloaded native library)
+      - name: Install parent + core net.ladenthin:llama (bundles the downloaded native library)
         run: >
-          mvn -B --no-transfer-progress -DskipTests -Denforcer.skip=true
+          mvn -B --no-transfer-progress -pl llama -am -DskipTests -Denforcer.skip=true
           -Dspotless.check.skip=true -Dspotbugs.skip=true
           -Dmaven.javadoc.skip=true -Dmaven.source.skip=true -Dgpg.skip=true install
       - name: Run llama-langchain4j model-backed integration tests (reused cached models)
@@ -242,7 +233,7 @@ jobs:
         id: tag
         shell: bash
         run: |
-          TAG=$(grep -oE 'GIT_TAG[[:space:]]+b[0-9]+' CMakeLists.txt | grep -oE 'b[0-9]+' | head -1)
+          TAG=$(grep -oE 'GIT_TAG[[:space:]]+b[0-9]+' llama/CMakeLists.txt | grep -oE 'b[0-9]+' | head -1)
           if [ -z "$TAG" ]; then
             echo "could not resolve llama.cpp GIT_TAG (b<nnnn>) from CMakeLists.txt" >&2
             exit 1
@@ -284,14 +275,14 @@ jobs:
             done )
           # llama-ui-embed is a self-contained C++17 host tool (no npm) — build + run it.
           g++ -O2 -std=c++17 -o llama-ui-embed embed.cpp
-          mkdir -p "$GITHUB_WORKSPACE/webui-generated"
+          mkdir -p "$GITHUB_WORKSPACE/llama/webui-generated"
           ./llama-ui-embed \
-            "$GITHUB_WORKSPACE/webui-generated/ui.cpp" \
-            "$GITHUB_WORKSPACE/webui-generated/ui.h" \
+            "$GITHUB_WORKSPACE/llama/webui-generated/ui.cpp" \
+            "$GITHUB_WORKSPACE/llama/webui-generated/ui.h" \
             dist
           echo "=== generated WebUI assets ==="
-          ls -la "$GITHUB_WORKSPACE/webui-generated"
-          if grep -q LLAMA_UI_HAS_ASSETS "$GITHUB_WORKSPACE/webui-generated/ui.h"; then
+          ls -la "$GITHUB_WORKSPACE/llama/webui-generated"
+          if grep -q LLAMA_UI_HAS_ASSETS "$GITHUB_WORKSPACE/llama/webui-generated/ui.h"; then
             echo "LLAMA_UI_HAS_ASSETS: present (real WebUI embedded)"
           else
             echo "ERROR: embed produced an empty asset table" >&2
@@ -301,7 +292,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
           retention-days: 1
           if-no-files-found: error
 
@@ -332,7 +323,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Display CPU Info
         shell: bash
         run: |
@@ -349,7 +340,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: linux-libraries-cuda
-          path: ${{ github.workspace }}/src/main/resources_linux_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_linux_cuda/net/ladenthin/llama/
 
   crosscompile-linux-x86_64:
     name: Cross-Compile manylinux2014 x86_64
@@ -372,7 +363,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Display CPU Info
         shell: bash
         run: |
@@ -389,7 +380,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: Linux-x86_64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   crosscompile-linux-aarch64:
     name: Build and Test Linux aarch64
@@ -414,7 +405,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - uses: actions/setup-java@v5
         with:
           distribution: 'temurin'
@@ -436,15 +427,15 @@ jobs:
       - name: Build libraries
         shell: bash
         run: |
-          mvn --no-transfer-progress compile
+          mvn --no-transfer-progress -f llama/pom.xml compile
           .github/build.sh "-DOS_NAME=Linux -DOS_ARCH=aarch64 -DGGML_NATIVE=OFF -DBUILD_TESTING=ON"
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: Linux-aarch64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   crosscompile-android-aarch64:
     name: Cross-Compile Android aarch64
@@ -464,7 +455,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Display CPU Info
         shell: bash
         run: |
@@ -481,7 +472,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: Linux-Android-aarch64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   crosscompile-android-aarch64-opencl:
     name: Cross-Compile Android aarch64 (OpenCL/Adreno)
@@ -502,7 +493,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Build libraries
         shell: bash
         run: |
@@ -511,7 +502,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: android-libraries-opencl
-          path: ${{ github.workspace }}/src/main/resources_android_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_android_opencl/net/ladenthin/llama/
 
   # ---------------------------------------------------------------------------
   # Native build jobs — produce release artifacts + run C++ unit tests
@@ -532,7 +523,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - uses: actions/setup-java@v5
         with:
           distribution: 'temurin'
@@ -552,15 +543,15 @@ jobs:
       - name: Build libraries
         shell: bash
         run: |
-          mvn --no-transfer-progress compile
+          mvn --no-transfer-progress -f llama/pom.xml compile
           .github/build.sh -DLLAMA_METAL=OFF -DGGML_NATIVE=OFF -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: macos-15-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   build-macos-arm64-metal:
     name: Build and Test macOS 14 arm64 (Metal)
@@ -577,7 +568,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - uses: actions/setup-java@v5
         with:
           distribution: 'temurin'
@@ -597,15 +588,15 @@ jobs:
       - name: Build libraries
         shell: bash
         run: |
-          mvn --no-transfer-progress compile
+          mvn --no-transfer-progress -f llama/pom.xml compile
           .github/build.sh -DLLAMA_METAL_EMBED_LIBRARY=ON -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: macos-14-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   build-windows-x86_64-msvc:
     name: Build and Test Windows 2025 x86_64 (MSVC / VS 2026, classifier)
@@ -617,7 +608,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Display CPU Info
         shell: pwsh
         run: |
@@ -634,12 +625,12 @@ jobs:
         run: |
           .github\build.bat -G "Visual Studio 18 2026" -A "x64" -DOS_NAME=Windows -DOS_ARCH=x86_64 -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86_64-msvc
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   build-windows-x86-msvc:
     name: Build and Test Windows 2025 x86 (MSVC / VS 2026, classifier)
@@ -651,7 +642,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Display CPU Info
         shell: pwsh
         run: |
@@ -668,12 +659,12 @@ jobs:
         run: |
           .github\build.bat -G "Visual Studio 18 2026" -A "Win32" -DOS_NAME=Windows -DOS_ARCH=x86 -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86-msvc
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   # ---------------------------------------------------------------------------
   # Windows Ninja Multi-Config + sccache — the DEFAULT Windows CPU natives.
@@ -703,7 +694,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Set up MSVC developer environment (x64)
         uses: ilammy/msvc-dev-cmd@v1
         with:
@@ -736,12 +727,12 @@ jobs:
         run: |
           .github\build.bat -G "Ninja Multi-Config" -DOS_NAME=Windows -DOS_ARCH=x86_64 -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86_64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   build-windows-x86:
     name: Build and Test Windows 2025 x86 (Ninja Multi-Config + sccache, default)
@@ -757,7 +748,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Set up MSVC developer environment (x86)
         uses: ilammy/msvc-dev-cmd@v1
         with:
@@ -790,12 +781,12 @@ jobs:
         run: |
           .github\build.bat -G "Ninja Multi-Config" -DOS_NAME=Windows -DOS_ARCH=x86 -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   # ---------------------------------------------------------------------------
   # Windows GPU classifiers (x86_64 only) — CUDA, Vulkan, OpenCL.
@@ -826,7 +817,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Set up MSVC developer environment (x64)
         uses: ilammy/msvc-dev-cmd@v1
         with:
@@ -864,7 +855,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86_64-cuda
-          path: ${{ github.workspace }}/src/main/resources_windows_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_cuda/net/ladenthin/llama/
           if-no-files-found: error
 
   build-windows-x86_64-vulkan:
@@ -881,7 +872,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Set up MSVC developer environment (x64)
         uses: ilammy/msvc-dev-cmd@v1
         with:
@@ -913,7 +904,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86_64-vulkan
-          path: ${{ github.workspace }}/src/main/resources_windows_vulkan/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_vulkan/net/ladenthin/llama/
           if-no-files-found: error
 
   build-windows-x86_64-opencl:
@@ -930,7 +921,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - name: Set up MSVC developer environment (x64)
         uses: ilammy/msvc-dev-cmd@v1
         with:
@@ -957,7 +948,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: Windows-x86_64-opencl
-          path: ${{ github.workspace }}/src/main/resources_windows_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_opencl/net/ladenthin/llama/
           if-no-files-found: error
 
   # ---------------------------------------------------------------------------
@@ -983,10 +974,10 @@ jobs:
           cat /proc/cpuinfo
       - name: Build libraries
         run: |
-          mvn -q --no-transfer-progress compile
+          mvn -q --no-transfer-progress -f llama/pom.xml compile
           .github/build.sh -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
 
   build-macos-arm64-metal-15:
     name: Build and Test macOS 15 arm64 (Metal)
@@ -1003,7 +994,7 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: webui-generated
-          path: ${{ github.workspace }}/webui-generated/
+          path: ${{ github.workspace }}/llama/webui-generated/
       - uses: actions/setup-java@v5
         with:
           distribution: 'temurin'
@@ -1023,15 +1014,15 @@ jobs:
       - name: Build libraries
         shell: bash
         run: |
-          mvn --no-transfer-progress compile
+          mvn --no-transfer-progress -f llama/pom.xml compile
           .github/build.sh -DLLAMA_METAL_EMBED_LIBRARY=ON -DGGML_NATIVE=OFF -DBUILD_TESTING=ON
       - name: Run C++ unit tests
-        run: ctest --test-dir build --output-on-failure
+        run: ctest --test-dir llama/build --output-on-failure
       - name: Upload artifacts
         uses: actions/upload-artifact@v7
         with:
           name: macos-15-metal-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
 
   # ---------------------------------------------------------------------------
   # Java test jobs — download release artifact, run mvn test
@@ -1054,7 +1045,7 @@ jobs:
       - uses: actions/download-artifact@v8
         with:
           name: Linux-x86_64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       # GGUF models are downloaded + cached ONCE by the upstream `download-models` job
       # (this job `needs:` it), so here we only RESTORE the shared cache — no per-job
       # download logic. GGUF is platform-independent, so ubuntu + macOS + Windows share
@@ -1079,7 +1070,7 @@ jobs:
           echo "${{ github.workspace }}/core.%e.%p" | sudo tee /proc/sys/kernel/core_pattern
       - name: Run tests
         run: |
-          mvn -e --no-transfer-progress -P jcstress test \
+          mvn -e --no-transfer-progress -f llama/pom.xml -P jcstress test \
             -Dnet.ladenthin.llama.tool.model=models/${TOOL_MODEL_NAME} \
             -Dnet.ladenthin.llama.nomic.path=models/${NOMIC_EMBED_MODEL_NAME} \
             -Dnet.ladenthin.llama.vision.model=models/${VISION_MODEL_NAME} \
@@ -1091,15 +1082,15 @@ jobs:
         if: success()
         with:
           name: jacoco-report
-          path: target/site/jacoco/jacoco.xml
+          path: llama/target/site/jacoco/jacoco.xml
           if-no-files-found: ignore
       - name: Run PIT mutation tests
-        run: mvn --batch-mode --no-transfer-progress test-compile org.pitest:pitest-maven:mutationCoverage
+        run: mvn --batch-mode --no-transfer-progress -f llama/pom.xml test-compile org.pitest:pitest-maven:mutationCoverage
       - name: Extract PIT survivors
         if: always()
         run: |
           echo "=== PIT Survived Mutations ==="
-          for html_file in $(find target/pit-reports -name "*.html" -type f 2>/dev/null | sort); do
+          for html_file in $(find llama/target/pit-reports -name "*.html" -type f 2>/dev/null | sort); do
             if grep -q "SURVIVED" "$html_file"; then
               echo "Found survivors in $html_file:"
               grep -B 2 -A 3 "SURVIVED" "$html_file"
@@ -1108,7 +1099,7 @@ jobs:
           done
       - uses: actions/upload-artifact@v7
         if: always()
-        with: { name: pit-reports, path: target/pit-reports/ }
+        with: { name: pit-reports, path: llama/target/pit-reports/ }
       - name: Memory after tests
         if: always()
         run: free -h
@@ -1146,13 +1137,13 @@ jobs:
         # -Dtest matches simple class names, not package globs; the default suite is
         # excluded from the vmlens package via pom.xml managed surefire <excludes>).
         run: >-
-          mvn --batch-mode --no-transfer-progress -Pvmlens test
+          mvn --batch-mode --no-transfer-progress -f llama/pom.xml -Pvmlens test
           -Dtest=VmlensInterleavingSmokeTest,SessionStateInterleavingTest -DfailIfNoTests=false
       - uses: actions/upload-artifact@v7
         if: always()
         with:
           name: vmlens-report
-          path: target/vmlens-report/
+          path: llama/target/vmlens-report/
           if-no-files-found: ignore
 
   test-java-macos-arm64-metal:
@@ -1172,7 +1163,7 @@ jobs:
       - uses: actions/download-artifact@v8
         with:
           name: macos-14-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       # GGUF models are downloaded + cached ONCE by the upstream `download-models` job
       # (this job `needs:` it), so here we only RESTORE the shared cache — no per-job
       # download logic. GGUF is platform-independent, so ubuntu + macOS + Windows share
@@ -1195,7 +1186,7 @@ jobs:
         run: ulimit -c unlimited
       - name: Run tests
         run: |
-          mvn -e --no-transfer-progress -Dnet.ladenthin.llama.test.ngl=0 test \
+          mvn -e --no-transfer-progress -f llama/pom.xml -Dnet.ladenthin.llama.test.ngl=0 test \
             -Dnet.ladenthin.llama.tool.model=models/${TOOL_MODEL_NAME} \
             -Dnet.ladenthin.llama.nomic.path=models/${NOMIC_EMBED_MODEL_NAME} \
             -Dnet.ladenthin.llama.vision.model=models/${VISION_MODEL_NAME} \
@@ -1236,7 +1227,7 @@ jobs:
       - uses: actions/download-artifact@v8
         with:
           name: macos-15-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       # GGUF models are downloaded + cached ONCE by the upstream `download-models` job
       # (this job `needs:` it), so here we only RESTORE the shared cache — no per-job
       # download logic. GGUF is platform-independent, so ubuntu + macOS + Windows share
@@ -1259,7 +1250,7 @@ jobs:
         run: ulimit -c unlimited
       - name: Run tests
         run: |
-          mvn -e --no-transfer-progress test \
+          mvn -e --no-transfer-progress -f llama/pom.xml test \
             -Dnet.ladenthin.llama.tool.model=models/${TOOL_MODEL_NAME} \
             -Dnet.ladenthin.llama.nomic.path=models/${NOMIC_EMBED_MODEL_NAME} \
             -Dnet.ladenthin.llama.vision.model=models/${VISION_MODEL_NAME} \
@@ -1300,7 +1291,7 @@ jobs:
       - uses: actions/download-artifact@v8
         with:
           name: macos-15-metal-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       # GGUF models are downloaded + cached ONCE by the upstream `download-models` job
       # (this job `needs:` it), so here we only RESTORE the shared cache — no per-job
       # download logic. GGUF is platform-independent, so ubuntu + macOS + Windows share
@@ -1323,7 +1314,7 @@ jobs:
         run: ulimit -c unlimited
       - name: Run tests
         run: |
-          mvn -e --no-transfer-progress test \
+          mvn -e --no-transfer-progress -f llama/pom.xml test \
             -Dnet.ladenthin.llama.tool.model=models/${TOOL_MODEL_NAME} \
             -Dnet.ladenthin.llama.nomic.path=models/${NOMIC_EMBED_MODEL_NAME} \
             -Dnet.ladenthin.llama.vision.model=models/${VISION_MODEL_NAME} \
@@ -1367,7 +1358,7 @@ jobs:
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-libraries
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       # GGUF models are downloaded + cached ONCE by the upstream `download-models` job
       # (this job `needs:` it), so here we only RESTORE the shared cache — no per-job
       # download logic. validate-models is kept as an integrity guard so a partial/absent
@@ -1405,7 +1396,7 @@ jobs:
           Get-ItemProperty -Path $key | Format-List
       - name: Run tests
         run: |
-          mvn -e --no-transfer-progress test `
+          mvn -e --no-transfer-progress -f llama/pom.xml test `
             "-Dnet.ladenthin.llama.tool.model=models/$env:TOOL_MODEL_NAME" `
             "-Dnet.ladenthin.llama.nomic.path=models/$env:NOMIC_EMBED_MODEL_NAME" `
             "-Dnet.ladenthin.llama.vision.model=models/$env:VISION_MODEL_NAME" `
@@ -1429,7 +1420,7 @@ jobs:
             ${{ github.workspace }}\target\surefire-reports\*.dumpstream
             ${{ github.workspace }}\target\surefire-reports\*.txt
             ${{ github.workspace }}\target\surefire-reports\TEST-*.xml
-            ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/**/*
+            ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/**/*
           if-no-files-found: warn
 
   # Java/inference validation of the MSVC-built x86_64 DLL (the analogue of
@@ -1456,7 +1447,7 @@ jobs:
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-msvc
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       # GGUF models are downloaded + cached ONCE by the upstream `download-models` job
       # (this job `needs:` it), so here we only RESTORE the shared cache — no per-job
       # download logic. validate-models is kept as an integrity guard so a partial/absent
@@ -1494,7 +1485,7 @@ jobs:
           Get-ItemProperty -Path $key | Format-List
       - name: Run tests
         run: |
-          mvn -e --no-transfer-progress test `
+          mvn -e --no-transfer-progress -f llama/pom.xml test `
             "-Dnet.ladenthin.llama.tool.model=models/$env:TOOL_MODEL_NAME" `
             "-Dnet.ladenthin.llama.nomic.path=models/$env:NOMIC_EMBED_MODEL_NAME" `
             "-Dnet.ladenthin.llama.vision.model=models/$env:VISION_MODEL_NAME" `
@@ -1518,7 +1509,7 @@ jobs:
             ${{ github.workspace }}\target\surefire-reports\*.dumpstream
             ${{ github.workspace }}\target\surefire-reports\*.txt
             ${{ github.workspace }}\target\surefire-reports\TEST-*.xml
-            ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/**/*
+            ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/**/*
           if-no-files-found: warn
 
   # ---------------------------------------------------------------------------
@@ -1554,38 +1545,38 @@ jobs:
         with:
           pattern: "*-libraries"
           merge-multiple: true
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: linux-libraries-cuda
-          path: ${{ github.workspace }}/src/main/resources_linux_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_linux_cuda/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: android-libraries-opencl
-          path: ${{ github.workspace }}/src/main/resources_android_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_android_opencl/net/ladenthin/llama/
       # MSVC-built Windows natives -> `msvc-windows` classifier tree. The default JAR
       # now ships the Ninja `*-libraries` natives merged above (default flip).
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-msvc
-          path: ${{ github.workspace }}/src/main/resources_windows_msvc/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_msvc/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86-msvc
-          path: ${{ github.workspace }}/src/main/resources_windows_msvc/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_msvc/net/ladenthin/llama/
       # Windows GPU classifiers (x86_64 only) -> one tree each.
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-cuda
-          path: ${{ github.workspace }}/src/main/resources_windows_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_cuda/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-vulkan
-          path: ${{ github.workspace }}/src/main/resources_windows_vulkan/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_vulkan/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-opencl
-          path: ${{ github.workspace }}/src/main/resources_windows_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_opencl/net/ladenthin/llama/
       - uses: actions/setup-java@v5
         with:
           distribution: 'temurin'
@@ -1604,7 +1595,7 @@ jobs:
         uses: actions/upload-artifact@v7
         with:
           name: llama-jars
-          path: target/*.jar
+          path: llama/target/*.jar
 
   report:
     name: Report
@@ -1668,35 +1659,35 @@ jobs:
         with:
           pattern: "*-libraries"
           merge-multiple: true
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: linux-libraries-cuda
-          path: ${{ github.workspace }}/src/main/resources_linux_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_linux_cuda/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: android-libraries-opencl
-          path: ${{ github.workspace }}/src/main/resources_android_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_android_opencl/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-msvc
-          path: ${{ github.workspace }}/src/main/resources_windows_msvc/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_msvc/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86-msvc
-          path: ${{ github.workspace }}/src/main/resources_windows_msvc/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_msvc/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-cuda
-          path: ${{ github.workspace }}/src/main/resources_windows_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_cuda/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-vulkan
-          path: ${{ github.workspace }}/src/main/resources_windows_vulkan/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_vulkan/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-opencl
-          path: ${{ github.workspace }}/src/main/resources_windows_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_opencl/net/ladenthin/llama/
       - name: Set up Maven Central Repository
         uses: actions/setup-java@v5
         with:
@@ -1716,27 +1707,21 @@ jobs:
             *-SNAPSHOT) echo "OK: -SNAPSHOT version, continuing snapshot deploy." ;;
             *) echo "::error::Refusing to publish non-SNAPSHOT version '$VERSION' from the snapshot job. Snapshot publishing requires a -SNAPSHOT version; releases go through the v* tag path."; exit 1 ;;
           esac
-      - name: Publish snapshot
+      # One reactor deploy publishes all three artifacts at the same version:
+      # net.ladenthin:llama-parent (the pom), :llama (the core jar + classifiers), and
+      # :llama-langchain4j. The `release` profile (GPG + Central Publishing) is inherited
+      # from the parent, so every module — including the parent pom — is signed.
+      - name: Publish snapshot (reactor - parent + llama + llama-langchain4j)
         run: mvn --batch-mode --no-transfer-progress -P release,cuda,opencl-android,windows-msvc,cuda-windows,vulkan-windows,opencl-windows -Dmaven.test.skip=true deploy
         env:
           MAVEN_USERNAME: ${{ secrets.CENTRAL_USERNAME }}
           MAVEN_PASSWORD: ${{ secrets.CENTRAL_TOKEN }}
           MAVEN_GPG_PASSPHRASE: ${{ secrets.GPG_PASSPHRASE }}
-      # Deploy the sibling llama-langchain4j at the same version. The core `deploy`
-      # above ran through `install`, so net.ladenthin:llama is in the local repo for
-      # the module to resolve. Standalone module (not in the reactor), so it is a
-      # separate deploy invocation.
-      - name: Publish snapshot (llama-langchain4j)
-        run: mvn --batch-mode --no-transfer-progress -f llama-langchain4j/pom.xml -P release -Dmaven.test.skip=true deploy
-        env:
-          MAVEN_USERNAME: ${{ secrets.CENTRAL_USERNAME }}
-          MAVEN_PASSWORD: ${{ secrets.CENTRAL_TOKEN }}
-          MAVEN_GPG_PASSPHRASE: ${{ secrets.GPG_PASSPHRASE }}
       - name: Collect signed artifacts
         run: |
           mkdir -p signed-snapshot-assets
-          cp target/*.jar signed-snapshot-assets/ 2>/dev/null || true
-          cp target/*.jar.asc signed-snapshot-assets/ 2>/dev/null || true
+          cp llama/target/*.jar signed-snapshot-assets/ 2>/dev/null || true
+          cp llama/target/*.jar.asc signed-snapshot-assets/ 2>/dev/null || true
           cp llama-langchain4j/target/*.jar signed-snapshot-assets/ 2>/dev/null || true
           cp llama-langchain4j/target/*.jar.asc signed-snapshot-assets/ 2>/dev/null || true
       - uses: actions/upload-artifact@v7
@@ -1784,35 +1769,35 @@ jobs:
         with:
           pattern: "*-libraries"
           merge-multiple: true
-          path: ${{ github.workspace }}/src/main/resources/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: linux-libraries-cuda
-          path: ${{ github.workspace }}/src/main/resources_linux_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_linux_cuda/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: android-libraries-opencl
-          path: ${{ github.workspace }}/src/main/resources_android_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_android_opencl/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-msvc
-          path: ${{ github.workspace }}/src/main/resources_windows_msvc/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_msvc/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86-msvc
-          path: ${{ github.workspace }}/src/main/resources_windows_msvc/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_msvc/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-cuda
-          path: ${{ github.workspace }}/src/main/resources_windows_cuda/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_cuda/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-vulkan
-          path: ${{ github.workspace }}/src/main/resources_windows_vulkan/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_vulkan/net/ladenthin/llama/
       - uses: actions/download-artifact@v8
         with:
           name: Windows-x86_64-opencl
-          path: ${{ github.workspace }}/src/main/resources_windows_opencl/net/ladenthin/llama/
+          path: ${{ github.workspace }}/llama/src/main/resources_windows_opencl/net/ladenthin/llama/
       - name: Set up Maven Central Repository
         uses: actions/setup-java@v5
         with:
@@ -1823,27 +1808,21 @@ jobs:
           server-password: MAVEN_PASSWORD
           gpg-private-key: ${{ secrets.GPG_PRIVATE_KEY }}
           gpg-passphrase: MAVEN_GPG_PASSPHRASE
-      - name: Publish release
+      # One reactor deploy publishes all three artifacts at the same version:
+      # net.ladenthin:llama-parent (the pom), :llama (the core jar + classifiers), and
+      # :llama-langchain4j. The `release` profile (GPG + Central Publishing) is inherited
+      # from the parent, so every module — including the parent pom — is signed.
+      - name: Publish release (reactor - parent + llama + llama-langchain4j)
         run: mvn --batch-mode --no-transfer-progress -P release,cuda,opencl-android,windows-msvc,cuda-windows,vulkan-windows,opencl-windows -Dmaven.test.skip=true deploy
         env:
           MAVEN_USERNAME: ${{ secrets.CENTRAL_USERNAME }}
           MAVEN_PASSWORD: ${{ secrets.CENTRAL_TOKEN }}
           MAVEN_GPG_PASSPHRASE: ${{ secrets.GPG_PASSPHRASE }}
-      # Deploy the sibling llama-langchain4j at the same version. The core `deploy`
-      # above ran through `install`, so net.ladenthin:llama is in the local repo for
-      # the module to resolve. Standalone module (not in the reactor), so it is a
-      # separate deploy invocation.
-      - name: Publish release (llama-langchain4j)
-        run: mvn --batch-mode --no-transfer-progress -f llama-langchain4j/pom.xml -P release -Dmaven.test.skip=true deploy
-        env:
-          MAVEN_USERNAME: ${{ secrets.CENTRAL_USERNAME }}
-          MAVEN_PASSWORD: ${{ secrets.CENTRAL_TOKEN }}
-          MAVEN_GPG_PASSPHRASE: ${{ secrets.GPG_PASSPHRASE }}
       - name: Collect signed artifacts
         run: |
           mkdir -p signed-release-assets
-          cp target/*.jar signed-release-assets/ 2>/dev/null || true
-          cp target/*.jar.asc signed-release-assets/ 2>/dev/null || true
+          cp llama/target/*.jar signed-release-assets/ 2>/dev/null || true
+          cp llama/target/*.jar.asc signed-release-assets/ 2>/dev/null || true
           cp llama-langchain4j/target/*.jar signed-release-assets/ 2>/dev/null || true
           cp llama-langchain4j/target/*.jar.asc signed-release-assets/ 2>/dev/null || true
       - uses: actions/upload-artifact@v7
diff --git a/.github/workflows/sonarqube.yml b/.github/workflows/sonarqube.yml
index b6a4808b..2ee66a0a 100644
--- a/.github/workflows/sonarqube.yml
+++ b/.github/workflows/sonarqube.yml
@@ -37,4 +37,4 @@ jobs:
       - name: Build and analyze
         env:
           SONAR_TOKEN: ${{ secrets.SONAR_TOKEN }}
-        run: mvn -B --no-transfer-progress verify org.sonarsource.scanner.maven:sonar-maven-plugin:sonar -Dsonar.projectKey=bernardladenthin_java-llama.cpp
+        run: mvn -B --no-transfer-progress -f llama/pom.xml verify org.sonarsource.scanner.maven:sonar-maven-plugin:sonar -Dsonar.projectKey=bernardladenthin_java-llama.cpp
diff --git a/.gitignore b/.gitignore
index be02ca4b..ba0a69ed 100644
--- a/.gitignore
+++ b/.gitignore
@@ -38,28 +38,28 @@ hs_err_pid*
 replay_pid*
 
 models/*.gguf
-src/main/cpp/net_ladenthin_llama_*.h
-src/main/resources_cuda_linux/
+llama/src/main/cpp/net_ladenthin_llama_*.h
+llama/src/main/resources_cuda_linux/
 # Per-classifier native trees, staged by CI before the matching Maven profile runs,
 # never committed (same policy as the default-tree native libs below).
-src/main/resources_windows_msvc/
-src/main/resources_windows_cuda/
-src/main/resources_windows_vulkan/
-src/main/resources_windows_opencl/
-src/main/resources/**/*.so
-src/main/resources/**/*.dylib
-src/main/resources/**/*.dll
-src/main/resources/**/*.metal
-src/test/resources/**/*.gbnf
+llama/src/main/resources_windows_msvc/
+llama/src/main/resources_windows_cuda/
+llama/src/main/resources_windows_vulkan/
+llama/src/main/resources_windows_opencl/
+llama/src/main/resources/**/*.so
+llama/src/main/resources/**/*.dylib
+llama/src/main/resources/**/*.dll
+llama/src/main/resources/**/*.metal
+llama/src/test/resources/**/*.gbnf
 
 # Generated WebUI assets (ui.cpp/ui.h) produced once by the build-webui CI job and
 # downloaded into every native build; embedded into libjllama, never committed
 # (this repo commits no build outputs — same policy as the native libs above).
-/webui-generated/
+/llama/webui-generated/
 
 **/*.etag
 **/*.lastModified
-src/main/cpp/llama.cpp/
+llama/src/main/cpp/llama.cpp/
 
 # jcstress / jqwik test outputs (generated in repo root)
 /.jqwik-database
diff --git a/CLAUDE.md b/CLAUDE.md
index db03010f..86d74121 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -16,7 +16,7 @@ To change the CUDA version, update the following **three** places:
 
 1. **`.github/build_cuda_linux.sh`** — Line 10: `sudo dnf install -y cuda-toolkit-13-2`
 2. **`.github/build_cuda_linux.sh`** — Line 12: `-DCMAKE_CUDA_COMPILER=/usr/local/cuda-13.2/bin/nvcc`
-3. **`pom.xml`** — The `<classifier>` tag in the `cuda` jar execution: `cuda13-linux-x86-64`
+3. **`llama/pom.xml`** — The `<classifier>` tag in the `cuda` jar execution: `cuda13-linux-x86-64`
 
 Also update the header comment in `build_cuda_linux.sh` and the job name in `.github/workflows/release.yaml` for clarity.
 
@@ -32,9 +32,9 @@ Example: To upgrade from 13.2 to a hypothetical 13.3:
 # Edit .github/build_cuda_linux.sh:
 #   line 10: cuda-toolkit-13-2 -> cuda-toolkit-13-3
 #   line 12: /usr/local/cuda-13.2/bin/nvcc -> /usr/local/cuda-13.3/bin/nvcc
-# Edit pom.xml classifier: cuda13-linux-x86-64 (major version only, no need to change for minor bumps)
+# Edit llama/pom.xml classifier: cuda13-linux-x86-64 (major version only, no need to change for minor bumps)
 # Edit CLAUDE.md line: Current CUDA version: **13.2** -> **13.3**
-git add .github/build_cuda_linux.sh pom.xml CLAUDE.md
+git add .github/build_cuda_linux.sh llama/pom.xml CLAUDE.md
 git commit -m "Upgrade CUDA from 13.2 to 13.3"
 ```
 
@@ -88,7 +88,7 @@ This is enforced through bionic's **weak-symbol** mechanism, *not* by bumping
 `__ANDROID_API__` or passing `-DANDROID_PLATFORM`. See "How the API gate is
 satisfied" below for why. To change anything here, update:
 
-1. **`CMakeLists.txt`** — the `add_compile_definitions(__ANDROID_UNAVAILABLE_SYMBOLS_ARE_WEAK__)`
+1. **`llama/CMakeLists.txt`** — the `add_compile_definitions(__ANDROID_UNAVAILABLE_SYMBOLS_ARE_WEAK__)`
    block and its Android-detection guard (`OS_NAME MATCHES "Android"` etc.).
 2. **`CLAUDE.md`** (this file) — the "Current Android minimum API level" line above.
 3. **`README.md`** — the minimum-API note (the `[!NOTE]` block near the Android
@@ -134,7 +134,7 @@ The default Android arm64 JAR remains CPU-only.
 
 Three places wire it together (mirrors the CUDA classifier pattern):
 
-1. **`CMakeLists.txt`** — `elseif(GGML_OPENCL)` branch routes artifacts to
+1. **`llama/CMakeLists.txt`** — `elseif(GGML_OPENCL)` branch routes artifacts to
    `src/main/resources_android_opencl/net/ladenthin/llama/${OS_NAME}/${OS_ARCH}/`.
 2. **`.github/workflows/publish.yml`** — `crosscompile-android-aarch64-opencl`
    job runs the dockcross-android-arm64 build with
@@ -142,7 +142,7 @@ Three places wire it together (mirrors the CUDA classifier pattern):
    and uploads as artifact `android-libraries-opencl`. The `package`,
    `publish-snapshot`, and `publish-release` jobs download it into
    `resources_android_opencl/` and activate the `opencl-android` Maven profile.
-3. **`pom.xml`** — the `opencl-android` profile produces a second JAR with
+3. **`llama/pom.xml`** — the `opencl-android` profile produces a second JAR with
    `<classifier>opencl-android-aarch64</classifier>` from the
    `${project.build.outputDirectory}_opencl_android` tree.
 
@@ -196,7 +196,7 @@ local / self-hosted.
 
 Wiring (mirrors the CUDA-Linux / OpenCL-Android classifier pattern):
 
-1. **`CMakeLists.txt`** — the `if(GGML_CUDA) … elseif(GGML_VULKAN) … elseif(GGML_OPENCL) … else()`
+1. **`llama/CMakeLists.txt`** — the `if(GGML_CUDA) … elseif(GGML_VULKAN) … elseif(GGML_OPENCL) … else()`
    chain is **OS-aware**: CUDA → `resources_windows_cuda` on Windows (else `resources_linux_cuda`),
    Vulkan → `resources_windows_vulkan`, OpenCL → `resources_windows_opencl` on Windows (else
    `resources_android_opencl`). The default CPU build (both generators) still emits to the canonical
@@ -225,7 +225,7 @@ Wiring (mirrors the CUDA-Linux / OpenCL-Android classifier pattern):
    The `package`, `publish-snapshot`, and `publish-release` jobs download each non-default artifact into
    its `src/main/resources_windows_{msvc,cuda,vulkan,opencl}/` tree and activate the
    `windows-msvc,cuda-windows,vulkan-windows,opencl-windows` Maven profiles.
-5. **`pom.xml`** — profiles `windows-msvc` / `cuda-windows` / `vulkan-windows` / `opencl-windows`,
+5. **`llama/pom.xml`** — profiles `windows-msvc` / `cuda-windows` / `vulkan-windows` / `opencl-windows`,
    each a separate compile pass + resource copy + classified jar (classifiers `msvc-windows` /
    `cuda13-windows-x86-64` / `vulkan-windows-x86-64` / `opencl-windows-x86-64`). Activated only in CI.
 6. **`README.md`** — the classifier table + dependency snippets in "Choosing the right classifier".
@@ -263,7 +263,7 @@ checked in (same policy as the native libs).
 Pipeline (`.github/workflows/publish.yml`):
 
 1. **`build-webui` job** (ubuntu — the *only* job that runs `npm`): resolves the
-   pinned `b<nnnn>` tag from `CMakeLists.txt`'s `GIT_TAG`, sparse-checks-out
+   pinned `b<nnnn>` tag from `llama/CMakeLists.txt`'s `GIT_TAG`, sparse-checks-out
    `ggml-org/llama.cpp@<tag>` `tools/ui`, runs the upstream Svelte build
    (`npm ci && npm run build`), gzips `dist/` into `dist/_gzip/` (LLAMA_UI_GZIP
    parity), builds the self-contained `llama-ui-embed` host tool (plain C++17, **no
@@ -472,16 +472,16 @@ re-verify the generator the same way you re-verify `patches/`.
 
 To change the llama.cpp version, update the following **three** files (and re-verify `patches/`):
 
-1. **CMakeLists.txt** — the `GIT_TAG` line for llama.cpp: `GIT_TAG        b8831`
+1. **llama/CMakeLists.txt** — the `GIT_TAG` line for llama.cpp: `GIT_TAG        b8831`
 2. **README.md** — the badge and link line with the version number
 3. **CLAUDE.md** — the "Current llama.cpp pinned version" line
 
 Example: To upgrade from b8808 to b8831:
 ```bash
-# Edit CMakeLists.txt: change GIT_TAG b8808 to b8831
+# Edit llama/CMakeLists.txt: change GIT_TAG b8808 to b8831
 # Edit README.md: change b8808 to b8831 (in both badge and link)
 # Edit CLAUDE.md: change b8808 to b8831
-git add CMakeLists.txt README.md CLAUDE.md
+git add llama/CMakeLists.txt README.md CLAUDE.md
 git commit -m "Upgrade llama.cpp from b8808 to b8831"
 git push -u origin <your-branch>
 ```
@@ -1225,17 +1225,46 @@ keeping it clear of the JPMS module-mode javadoc trap that bit BAF. **Before rai
 javadoc source level to ≥ 9, read**
 [`../workspace/policies/jpms-module-descriptor.md`](../workspace/policies/jpms-module-descriptor.md).
 
-## LangChain4j integration (`llama-langchain4j` sibling module)
+## Repository layout — Maven reactor (`llama/` + `llama-langchain4j/`)
+
+The repo root is a thin **aggregator/parent POM** (`net.ladenthin:llama-parent`,
+`packaging=pom`) with two modules:
+
+- **`llama/`** — the native JNI core (`net.ladenthin:llama`). *All the core sources and build
+  files live here now:* `llama/src/`, `llama/CMakeLists.txt`, `llama/cmake/`, `llama/patches/`,
+  `llama/pom.xml`, `llama/spotbugs-exclude.xml`, `llama/lombok.config`, `llama/.clang-format`.
+  Its published coordinates are unchanged (`net.ladenthin:llama`), so consumers are unaffected.
+- **`llama-langchain4j/`** — the LangChain4j adapters (see below).
+
+Both modules inherit the single `<version>` from the parent, so they **ship in lockstep by
+construction** (no CI guard needed). The parent also holds the shared `release` profile (GPG +
+Central Publishing), so one reactor `mvn -P release deploy` signs and publishes all three
+artifacts (`llama-parent` pom, `llama`, `llama-langchain4j`) at the same version.
+
+**Consequences for build commands:** the core's cmake/native build runs *in `llama/`*.
+`.github/build.sh` / `build.bat` `cd` into `llama/` themselves (relative to the script), so CI
+and the dockcross containers (whose workdir stays the repo root) are unaffected. Locally, run
+core cmake builds from `llama/` (e.g. `cd llama && cmake -B build && cmake --build build`), and
+target the core with Maven via `-f llama/pom.xml` (or `-pl llama -am` from the root). A plain
+`mvn` at the root builds the whole reactor. **When a build-command example elsewhere in this
+file shows `cmake -B build` / `src/main/...` / `mvn compile` at the root, read it as running in
+`llama/`** (the paths moved; the recipes are otherwise unchanged).
+
+**Version bump:** change the `<version>` in the **root** `pom.xml` only; `llama` and
+`llama-langchain4j` inherit it. (The SNAPSHOT/`-SNAPSHOT` line and the README badge still need
+the usual manual update.)
+
+## LangChain4j integration (`llama-langchain4j` reactor module)
 
 `llama-langchain4j/` adapts a `LlamaModel` to LangChain4j's `ChatModel`,
 `StreamingChatModel`, `EmbeddingModel` and `ScoringModel` interfaces **in-process over
-JNI** (no HTTP hop). It is a **standalone sibling module**, deliberately *not* in the root
-reactor, so the native build/release pipeline is untouched.
+JNI** (no HTTP hop). It is a **reactor module** alongside the core `llama` module (see
+"Repository layout" above), so it is built, versioned and released together with the core.
 
 Why it is a **separate artifact** and not a classifier of the core: langchain4j 1.x
 requires **Java 17** (the core stays Java 8), and classifiers share the core's single POM —
 adding `langchain4j-core` there would force it (and the Java 17 floor) on every plain
-`net.ladenthin:llama` consumer. A separate `artifactId` with its own POM is the only way to
+`net.ladenthin:llama` consumer. A separate `artifactId` (its own module POM) is the only way to
 keep that dependency (and Java floor) off the core. It is pure Java with **no per-classifier
 matrix**: it compiles against the core's Java API, which is identical across every native
 classifier; the backend (CPU/CUDA/OpenCL/Vulkan) is a runtime classpath choice for the
@@ -1243,33 +1272,29 @@ consumer.
 
 Wiring:
 
-1. **`llama-langchain4j/pom.xml`** — `net.ladenthin:llama-langchain4j`, `release 17`,
-   depends on `net.ladenthin:llama:${project.version}` (so the core dep always matches the
-   module's own version) and `dev.langchain4j:langchain4j-core`. Carries its own
-   sources/javadoc/gpg + `release` profile (Central requires per-artifact signing; the module
-   has no parent to inherit them from — plugin versions are pinned in lockstep with the root
-   `pom.xml`). Java package stays `net.ladenthin.llama.langchain4j` (package name need not track
-   the artifactId).
-2. **`.github/workflows/publish.yml`** — the `test-java-llama-langchain4j` job installs the
-   core Java jar, runs a **version-lockstep guard** (module version must equal core version,
-   else the build fails — the standalone module can't inherit `${project.version}` from a
-   reactor), then `mvn -f llama-langchain4j/pom.xml verify` (7 model-free mapping unit tests
-   run; the 4 model-backed integration tests self-skip without a GGUF; `verify` also builds the
-   javadoc jar so a release-time javadoc break is caught in PR CI). The
-   `publish-snapshot`/`publish-release` jobs `needs:` this job and, after the core `deploy`
-   (which installs the core jar locally), run a second `deploy` of the module at the same
-   version. A separate **`test-java-llama-langchain4j-integration`** job runs the model-backed
-   tests (chat/streaming/embedding/scoring adapters) by **reusing** the shared GGUF cache
+1. **`llama-langchain4j/pom.xml`** — `net.ladenthin:llama-langchain4j`, `release 17`, a child of
+   `net.ladenthin:llama-parent` (so it **inherits `${project.version}`** — no hardcoded version,
+   no lockstep guard). Depends on `net.ladenthin:llama:${project.version}` and
+   `dev.langchain4j:langchain4j-core`. Builds its own sources/javadoc jars; the `release`
+   profile (GPG + Central Publishing) is **inherited from the parent**, not duplicated here.
+   Java package stays `net.ladenthin.llama.langchain4j` (package name need not track the artifactId).
+2. **`.github/workflows/publish.yml`** — the `test-java-llama-langchain4j` job installs
+   parent + core into the local repo (`mvn -pl llama -am -DskipTests install`), then
+   `mvn -f llama-langchain4j/pom.xml verify` (7 model-free mapping unit tests run; the 4
+   model-backed integration tests self-skip without a GGUF; `verify` also builds the javadoc
+   jar so a release-time javadoc break is caught in PR CI). The `publish-snapshot`/
+   `publish-release` jobs `needs:` this job; deployment is a **single reactor**
+   `mvn -P release deploy` (no separate module deploy step — the parent's inherited `release`
+   profile signs and publishes parent + llama + llama-langchain4j together at the same version).
+   A separate **`test-java-llama-langchain4j-integration`** job runs the model-backed tests
+   (chat/streaming/embedding/scoring adapters) by **reusing** the shared GGUF cache
    (`gguf-models-v1`, restore-only — no extra download) and the `Linux-x86_64-libraries` native
    artifact: it `needs: [crosscompile-linux-x86_64, download-models]` (so the cache is already
-   populated and it runs in parallel), installs the core jar with the downloaded native lib
-   bundled, and passes the already-cached chat
-   (`REASONING_MODEL_NAME`), nomic-embedding and jina-reranker model paths via the module's
+   populated and it runs in parallel), installs parent+core with the downloaded native lib
+   bundled, and passes the already-cached chat (`REASONING_MODEL_NAME`), nomic-embedding and
+   jina-reranker model paths via the module's
    `-Dnet.ladenthin.llama.langchain4j.{embedding,rerank}.model` / `net.ladenthin.llama.model.path`
    properties. It is validation-only (not a release gate); a cold cache degrades to a self-skip.
-3. **Version bumps** — when the root `pom.xml` `<version>` changes, bump
-   `llama-langchain4j/pom.xml` `<version>` to match in the same commit, or the lockstep guard
-   reds CI.
 
 **Open follow-ups** (documented in `llama-langchain4j/README.md`): tool calling
 (`ToolSpecification` ↔ jllama `ToolDefinition`), `response_format`/JSON mode, and multimodal
diff --git a/REUSE.toml b/REUSE.toml
index df4c5422..55720a91 100644
--- a/REUSE.toml
+++ b/REUSE.toml
@@ -19,7 +19,7 @@ path = [
     "docs/history/49be664_open_issues_comments.md",
     "docs/history/CHAT_INTEGRATION_SUMMARY.md",
     "docs/history/REFACTORING.md",
-    "src/test/resources/images/README.md",
+    "llama/src/test/resources/images/README.md",
     ".github/PULL_REQUEST_TEMPLATE.md",
     ".github/ISSUE_TEMPLATE/bug_report.md",
     ".github/ISSUE_TEMPLATE/feature_request.md",
@@ -43,7 +43,7 @@ SPDX-License-Identifier = "MIT"
 
 # CMakeLists.txt (CMake format not recognized by reuse annotate)
 [[annotations]]
-path = "CMakeLists.txt"
+path = "llama/CMakeLists.txt"
 SPDX-FileCopyrightText = [
     "2023-2025 Konstantin Herud",
     "2026 Bernard Ladenthin <bernard.ladenthin@gmail.com>",
@@ -79,13 +79,13 @@ SPDX-License-Identifier = "MIT"
 # diff). Glob covers every current and future patch dropped into patches/ — see CLAUDE.md
 # "Local llama.cpp source patches".
 [[annotations]]
-path = "patches/**"
+path = "llama/patches/**"
 SPDX-FileCopyrightText = "2026 Bernard Ladenthin <bernard.ladenthin@gmail.com>"
 SPDX-License-Identifier = "MIT"
 
 # Test image (binary, cannot carry inline SPDX)
 [[annotations]]
-path = "src/test/resources/images/test-image.jpg"
+path = "llama/src/test/resources/images/test-image.jpg"
 SPDX-FileCopyrightText = "2026 Bernard Ladenthin <bernard.ladenthin@gmail.com>"
 SPDX-License-Identifier = "MIT"
 
diff --git a/llama-langchain4j/pom.xml b/llama-langchain4j/pom.xml
index ecb0b8bb..dabee5b2 100644
--- a/llama-langchain4j/pom.xml
+++ b/llama-langchain4j/pom.xml
@@ -9,9 +9,14 @@ SPDX-License-Identifier: MIT
 	xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
 	<modelVersion>4.0.0</modelVersion>
 
-	<groupId>net.ladenthin</groupId>
+	<parent>
+		<groupId>net.ladenthin</groupId>
+		<artifactId>llama-parent</artifactId>
+		<version>5.0.4-SNAPSHOT</version>
+		<relativePath>../pom.xml</relativePath>
+	</parent>
+
 	<artifactId>llama-langchain4j</artifactId>
-	<version>5.0.4-SNAPSHOT</version>
 	<packaging>jar</packaging>
 
 	<name>${project.groupId}:${project.artifactId}</name>
@@ -61,8 +66,6 @@ SPDX-License-Identifier: MIT
 		<surefire.version>3.5.6</surefire.version>
 		<source.plugin.version>3.4.0</source.plugin.version>
 		<javadoc.plugin.version>3.12.0</javadoc.plugin.version>
-		<gpg.plugin.version>3.2.8</gpg.plugin.version>
-		<central.plugin.version>0.11.0</central.plugin.version>
 	</properties>
 
 	<dependencies>
@@ -147,47 +150,9 @@ SPDX-License-Identifier: MIT
 		</plugins>
 	</build>
 
-	<profiles>
-		<!-- Mirrors the core pom.xml `release` profile: GPG-sign the artifacts and route the
-		     deploy through the Central Publishing plugin. Activated by the publish-snapshot /
-		     publish-release CI jobs (`-P release ... deploy`). -->
-		<profile>
-			<id>release</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-gpg-plugin</artifactId>
-						<version>${gpg.plugin.version}</version>
-						<executions>
-							<execution>
-								<id>sign-artifacts</id>
-								<phase>verify</phase>
-								<goals><goal>sign</goal></goals>
-								<configuration>
-									<keyname>${gpg.keyname}</keyname>
-									<gpgArguments>
-										<arg>--pinentry-mode</arg>
-										<arg>loopback</arg>
-									</gpgArguments>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-
-					<plugin>
-						<groupId>org.sonatype.central</groupId>
-						<artifactId>central-publishing-maven-plugin</artifactId>
-						<version>${central.plugin.version}</version>
-						<extensions>true</extensions>
-						<configuration>
-							<publishingServerId>central</publishingServerId>
-							<autoPublish>true</autoPublish>
-							<waitUntil>published</waitUntil>
-						</configuration>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-	</profiles>
+	<!--
+		No `release` profile here: GPG signing + Central Publishing are inherited from the
+		parent pom (../pom.xml), so a single reactor `mvn -P release deploy` signs and
+		publishes this module alongside the core.
+	-->
 </project>
diff --git a/.clang-format b/llama/.clang-format
similarity index 100%
rename from .clang-format
rename to llama/.clang-format
diff --git a/.clang-tidy b/llama/.clang-tidy
similarity index 100%
rename from .clang-tidy
rename to llama/.clang-tidy
diff --git a/CMakeLists.txt b/llama/CMakeLists.txt
similarity index 98%
rename from CMakeLists.txt
rename to llama/CMakeLists.txt
index b687c1c1..51bebdfd 100644
--- a/CMakeLists.txt
+++ b/llama/CMakeLists.txt
@@ -277,10 +277,13 @@ endif()
 
 # include jni.h and jni_md.h
 if(NOT DEFINED JNI_INCLUDE_DIRS)
+    # The bundled JNI headers live in .github/include/ at the REPO ROOT, one level above
+    # this module (the core moved into llama/). Anchor to the parent of the CMake source
+    # dir so the path is correct regardless of the build CWD.
     if(OS_NAME MATCHES "^Linux" OR OS_NAME STREQUAL "Mac" OR OS_NAME STREQUAL "Darwin")
-        set(JNI_INCLUDE_DIRS .github/include/unix)
+        set(JNI_INCLUDE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/../.github/include/unix)
     elseif(OS_NAME STREQUAL "Windows")
-        set(JNI_INCLUDE_DIRS .github/include/windows)
+        set(JNI_INCLUDE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/../.github/include/windows)
     # if we don't have provided headers, try to find them via Java
     else()
         find_package(Java REQUIRED)
diff --git a/cmake/apply-llama-patches.cmake b/llama/cmake/apply-llama-patches.cmake
similarity index 100%
rename from cmake/apply-llama-patches.cmake
rename to llama/cmake/apply-llama-patches.cmake
diff --git a/cmake/generate-tts-upstream.cmake b/llama/cmake/generate-tts-upstream.cmake
similarity index 100%
rename from cmake/generate-tts-upstream.cmake
rename to llama/cmake/generate-tts-upstream.cmake
diff --git a/lombok.config b/llama/lombok.config
similarity index 100%
rename from lombok.config
rename to llama/lombok.config
diff --git a/patches/0001-win32-arg-parse-embed-guard.patch b/llama/patches/0001-win32-arg-parse-embed-guard.patch
similarity index 100%
rename from patches/0001-win32-arg-parse-embed-guard.patch
rename to llama/patches/0001-win32-arg-parse-embed-guard.patch
diff --git a/patches/0002-server-preserve-caller-load-progress-callback.patch b/llama/patches/0002-server-preserve-caller-load-progress-callback.patch
similarity index 100%
rename from patches/0002-server-preserve-caller-load-progress-callback.patch
rename to llama/patches/0002-server-preserve-caller-load-progress-callback.patch
diff --git a/patches/0003-pr22393-server-add-slot-prompt-similarity-getter-setter.patch b/llama/patches/0003-pr22393-server-add-slot-prompt-similarity-getter-setter.patch
similarity index 100%
rename from patches/0003-pr22393-server-add-slot-prompt-similarity-getter-setter.patch
rename to llama/patches/0003-pr22393-server-add-slot-prompt-similarity-getter-setter.patch
diff --git a/patches/0004-pr23116-server-per-request-reasoning-budget-tokens.patch b/llama/patches/0004-pr23116-server-per-request-reasoning-budget-tokens.patch
similarity index 100%
rename from patches/0004-pr23116-server-per-request-reasoning-budget-tokens.patch
rename to llama/patches/0004-pr23116-server-per-request-reasoning-budget-tokens.patch
diff --git a/llama/pom.xml b/llama/pom.xml
new file mode 100644
index 00000000..73b91a09
--- /dev/null
+++ b/llama/pom.xml
@@ -0,0 +1,1323 @@
+<!--
+SPDX-FileCopyrightText: 2026 Bernard Ladenthin <bernard.ladenthin@gmail.com>
+SPDX-FileCopyrightText: 2023-2025 Konstantin Herud
+
+SPDX-License-Identifier: MIT
+-->
+
+<project xmlns="http://maven.apache.org/POM/4.0.0"
+	xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+	xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
+	<modelVersion>4.0.0</modelVersion>
+
+	<parent>
+		<groupId>net.ladenthin</groupId>
+		<artifactId>llama-parent</artifactId>
+		<version>5.0.4-SNAPSHOT</version>
+		<relativePath>../pom.xml</relativePath>
+	</parent>
+
+	<artifactId>llama</artifactId>
+	<packaging>jar</packaging>
+
+	<name>${project.groupId}:${project.artifactId}</name>
+	<description>Java Bindings for llama.cpp - A Port of Facebook's LLaMA model
+		in C/C++.</description>
+	<url>https://github.com/bernardladenthin/java-llama.cpp</url>
+
+	<licenses>
+		<license>
+			<name>MIT License</name>
+			<url>https://www.opensource.org/licenses/mit-license.php</url>
+			<distribution>repo</distribution>
+		</license>
+	</licenses>
+
+	<developers>
+		<developer>
+			<name>Bernard Ladenthin</name>
+			<organizationUrl>https://github.com/bernardladenthin</organizationUrl>
+		</developer>
+	</developers>
+
+	<scm>
+		<connection>scm:git:https://github.com/bernardladenthin/java-llama.cpp.git</connection>
+		<developerConnection>scm:git:https://github.com/bernardladenthin/java-llama.cpp.git</developerConnection>
+		<url>https://github.com/bernardladenthin/java-llama.cpp/tree/main</url>
+	</scm>
+
+	<distributionManagement>
+		<snapshotRepository>
+			<id>central</id>
+			<url>https://central.sonatype.com/repository/maven-snapshots/</url>
+		</snapshotRepository>
+	</distributionManagement>
+
+	<properties>
+		<sonar.organization>bernardladenthin</sonar.organization>
+		<jna.version>5.19.0</jna.version>
+		<jspecify.version>1.0.0</jspecify.version>
+		<lombok.version>1.18.46</lombok.version>
+		<errorprone.version>2.50.0</errorprone.version>
+		<nullaway.version>0.13.7</nullaway.version>
+		<checker.version>4.2.0</checker.version>
+		<jackson.version>2.22.0</jackson.version>
+		<reactor.version>3.8.6</reactor.version>
+		<slf4j.version>2.0.18</slf4j.version>
+		<logback.version>1.5.37</logback.version>
+		<animal-sniffer.version>1.27</animal-sniffer.version>
+		<junit.version>6.1.1</junit.version>
+		<hamcrest.version>3.0</hamcrest.version>
+		<jmh.version>1.37</jmh.version>
+		<jcstress.version>0.16</jcstress.version>
+		<lincheck.version>3.6</lincheck.version>
+		<logcaptor.version>2.12.6</logcaptor.version>
+		<vmlens.version>1.2.28</vmlens.version>
+		<!-- DO NOT UPGRADE jqwik past 1.9.3. jqwik 1.10.0 added a deliberate
+		     anti-AI prompt-injection string to test stdout; the 1.10.1 user
+		     guide states the library "is not meant to be used by any 'AI'
+		     coding agents at all." 1.9.3 is the last pre-disclosure release
+		     and is the pinned version for this repo. Any CI / Dependabot /
+		     contributor PR that bumps this MUST be rejected. See CLAUDE.md
+		     section "jqwik prompt-injection in test output" for full context. -->
+		<jqwik.version>1.9.3</jqwik.version>
+		<archunit.version>1.4.2</archunit.version>
+		<spotbugs.version>4.10.2.0</spotbugs.version>
+		<fb-contrib.version>7.7.4</fb-contrib.version>
+		<findsecbugs.version>1.14.0</findsecbugs.version>
+		<spotless.version>3.7.0</spotless.version>
+		<palantir-java-format.version>2.94.0</palantir-java-format.version>
+		<project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
+		<project.build.outputTimestamp>${git.commit.time}</project.build.outputTimestamp>
+	</properties>
+
+	<!--
+	    Pin transitively-conflicting versions so maven-enforcer's
+	    dependencyConvergence rule passes. The direct slf4j-api version above
+	    must always win over the older 2.0.17 that logback-classic 1.5.32
+	    brings in transitively.
+	-->
+	<dependencyManagement>
+		<dependencies>
+			<dependency>
+				<groupId>org.slf4j</groupId>
+				<artifactId>slf4j-api</artifactId>
+				<version>${slf4j.version}</version>
+			</dependency>
+		</dependencies>
+	</dependencyManagement>
+
+	<dependencies>
+		<dependency>
+			<groupId>org.projectlombok</groupId>
+			<artifactId>lombok</artifactId>
+			<version>${lombok.version}</version>
+			<scope>provided</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.junit.jupiter</groupId>
+			<artifactId>junit-jupiter</artifactId>
+			<version>${junit.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.hamcrest</groupId>
+			<artifactId>hamcrest</artifactId>
+			<version>${hamcrest.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>net.jqwik</groupId>
+			<artifactId>jqwik</artifactId>
+			<version>${jqwik.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>com.tngtech.archunit</groupId>
+			<artifactId>archunit-junit5</artifactId>
+			<version>${archunit.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.jspecify</groupId>
+			<artifactId>jspecify</artifactId>
+			<version>${jspecify.version}</version>
+		</dependency>
+		<dependency>
+			<groupId>org.checkerframework</groupId>
+			<artifactId>checker-qual</artifactId>
+			<version>${checker.version}</version>
+		</dependency>
+		<dependency>
+			<groupId>com.fasterxml.jackson.core</groupId>
+			<artifactId>jackson-databind</artifactId>
+			<version>${jackson.version}</version>
+		</dependency>
+		<!-- Required by OSInfo (vendored from xerial/sqlite-jdbc) for log emission. -->
+		<dependency>
+			<groupId>org.slf4j</groupId>
+			<artifactId>slf4j-api</artifactId>
+			<version>${slf4j.version}</version>
+		</dependency>
+		<!-- Default SLF4J binding shipped with this library. Runtime scope: not
+		     required on the compile classpath, only loaded at JVM startup. -->
+		<dependency>
+			<groupId>ch.qos.logback</groupId>
+			<artifactId>logback-classic</artifactId>
+			<version>${logback.version}</version>
+			<scope>runtime</scope>
+		</dependency>
+		<!-- @IgnoreJRERequirement marker used by OSInfo (vendored from xerial/sqlite-jdbc)
+		     on a few methods that touch APIs unavailable on Android. Retention=CLASS, so
+		     the dep is compile-only (provided scope). -->
+		<dependency>
+			<groupId>org.codehaus.mojo</groupId>
+			<artifactId>animal-sniffer-annotations</artifactId>
+			<version>${animal-sniffer.version}</version>
+			<scope>provided</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.openjdk.jmh</groupId>
+			<artifactId>jmh-core</artifactId>
+			<version>${jmh.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.openjdk.jmh</groupId>
+			<artifactId>jmh-generator-annprocess</artifactId>
+			<version>${jmh.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.openjdk.jcstress</groupId>
+			<artifactId>jcstress-core</artifactId>
+			<version>${jcstress.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.jetbrains.lincheck</groupId>
+			<artifactId>lincheck</artifactId>
+			<version>${lincheck.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<!-- Test-only SLF4J/Logback log-capture for the LoggingSmokeTest. -->
+		<dependency>
+			<groupId>io.github.hakky54</groupId>
+			<artifactId>logcaptor</artifactId>
+			<version>${logcaptor.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<!-- Test-only Project Reactor used by ReactorIntegrationTest to prove the
+		     "wrap LlamaIterable with Flux.fromIterable + subscribeOn" pattern works
+		     end-to-end (backpressure, AutoCloseable cancel propagation). NOT a runtime
+		     dependency — consumers bring whichever reactive lib they already use
+		     (Reactor / RxJava / Kotlin Flow / Akka — see README "Reactive integration"). -->
+		<dependency>
+			<groupId>io.projectreactor</groupId>
+			<artifactId>reactor-core</artifactId>
+			<version>${reactor.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>io.projectreactor</groupId>
+			<artifactId>reactor-test</artifactId>
+			<version>${reactor.version}</version>
+			<scope>test</scope>
+		</dependency>
+		<!--
+			vmlens interleaving-analysis API. Test-scoped and transitive-dependency-free
+			(its own deps are all test-scope, hence not propagated), so it is safe on the
+			default test classpath under dependencyConvergence. Needed so the
+			VmlensInterleavingSmoke* test compiles in every build; it is only *executed*
+			under the `vmlens` profile (see that profile + the vmlens CI job). The default
+			surefire run excludes it (managed surefire <excludes> below) so the vmlens
+			agent's JDK 9 API class is never loaded outside the agent-driven run.
+		-->
+		<dependency>
+			<groupId>com.vmlens</groupId>
+			<artifactId>api</artifactId>
+			<version>${vmlens.version}</version>
+			<scope>test</scope>
+		</dependency>
+	</dependencies>
+
+	<build>
+		<pluginManagement>
+			<plugins>
+				<plugin>
+					<groupId>com.diffplug.spotless</groupId>
+					<artifactId>spotless-maven-plugin</artifactId>
+					<version>${spotless.version}</version>
+				</plugin>
+				<plugin>
+					<groupId>com.github.spotbugs</groupId>
+					<artifactId>spotbugs-maven-plugin</artifactId>
+					<version>${spotbugs.version}</version>
+				</plugin>
+				<plugin>
+					<groupId>com.vmlens</groupId>
+					<artifactId>vmlens-maven-plugin</artifactId>
+					<version>${vmlens.version}</version>
+				</plugin>
+				<plugin>
+					<groupId>io.github.git-commit-id</groupId>
+					<artifactId>git-commit-id-maven-plugin</artifactId>
+					<version>10.0.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-assembly-plugin</artifactId>
+					<version>3.8.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-compiler-plugin</artifactId>
+					<version>3.15.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-gpg-plugin</artifactId>
+					<version>3.2.8</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-jar-plugin</artifactId>
+					<version>3.5.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-javadoc-plugin</artifactId>
+					<version>3.12.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-resources-plugin</artifactId>
+					<version>3.5.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-source-plugin</artifactId>
+					<version>3.4.0</version>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-surefire-plugin</artifactId>
+					<version>3.5.6</version>
+					<configuration>
+						<!--
+							Tests in the `vmlens` package are meaningful only when run
+							under the vmlens agent (the `vmlens` profile / CI job). Without the
+							agent its AllInterleavings loop body never executes (a vacuous pass
+							that also prints an "agent not configured" warning), so exclude it
+							from the ordinary surefire run. The vmlens job re-includes them with
+							`-Dtest=...`, which overrides this exclude.
+						-->
+						<excludes>
+							<exclude>**/vmlens/*.java</exclude>
+						</excludes>
+					</configuration>
+				</plugin>
+				<plugin>
+					<groupId>org.apache.maven.plugins</groupId>
+					<artifactId>maven-enforcer-plugin</artifactId>
+					<version>3.6.3</version>
+				</plugin>
+				<plugin>
+					<groupId>org.codehaus.mojo</groupId>
+					<artifactId>exec-maven-plugin</artifactId>
+					<version>3.6.3</version>
+				</plugin>
+				<plugin>
+					<groupId>org.jacoco</groupId>
+					<artifactId>jacoco-maven-plugin</artifactId>
+					<version>0.8.15</version>
+				</plugin>
+				<plugin>
+					<groupId>org.pitest</groupId>
+					<artifactId>pitest-maven</artifactId>
+					<version>1.25.5</version>
+				</plugin>
+				<plugin>
+					<groupId>org.sonatype.central</groupId>
+					<artifactId>central-publishing-maven-plugin</artifactId>
+					<version>0.11.0</version>
+				</plugin>
+			</plugins>
+		</pluginManagement>
+		<plugins>
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-enforcer-plugin</artifactId>
+				<executions>
+					<execution>
+						<id>enforce</id>
+						<goals>
+							<goal>enforce</goal>
+						</goals>
+						<configuration>
+							<rules>
+								<requireMavenVersion>
+									<version>[3.6.3,)</version>
+								</requireMavenVersion>
+								<requireJavaVersion>
+									<version>[1.8,)</version>
+								</requireJavaVersion>
+								<dependencyConvergence/>
+								<bannedDependencies>
+									<excludes>
+										<!-- Use SLF4J directly; never commons-logging. -->
+										<exclude>commons-logging:commons-logging</exclude>
+										<!-- Legacy / EoL logging frameworks. -->
+										<exclude>log4j:log4j</exclude>
+										<!-- Old hamcrest split artifacts; use hamcrest:hamcrest 2.x+. -->
+										<exclude>org.hamcrest:hamcrest-core</exclude>
+										<exclude>org.hamcrest:hamcrest-library</exclude>
+										<exclude>org.hamcrest:hamcrest-all</exclude>
+										<!-- Legacy JUnit 3 / 4; we use Jupiter (junit:junit-jupiter). -->
+										<exclude>junit:junit</exclude>
+										<exclude>junit:junit-dep</exclude>
+									</excludes>
+								</bannedDependencies>
+							</rules>
+						</configuration>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<groupId>io.github.git-commit-id</groupId>
+				<artifactId>git-commit-id-maven-plugin</artifactId>
+				<executions>
+					<execution>
+						<id>get-git-properties</id>
+						<goals>
+							<goal>revision</goal>
+						</goals>
+						<phase>initialize</phase>
+					</execution>
+				</executions>
+				<configuration>
+					<dateFormat>yyyy-MM-dd'T'HH:mm:ss'Z'</dateFormat>
+					<dateFormatTimeZone>UTC</dateFormatTimeZone>
+					<generateGitPropertiesFile>false</generateGitPropertiesFile>
+					<failOnNoGitDirectory>false</failOnNoGitDirectory>
+				</configuration>
+			</plugin>
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-compiler-plugin</artifactId>
+				<configuration>
+					<release>8</release>
+					<testRelease>21</testRelease>
+					<showWarnings>true</showWarnings>
+					<parameters>true</parameters>
+					<compilerArgs>
+						<!--
+						  -Xlint:all surfaces every category of javac warning. Excluded:
+						    -serial    : serialVersionUID warnings on non-Serializable classes
+						    -options   : bootclasspath / target-mismatch chatter under release=8
+						    -classfile : noise when javac reads our own .class output mid-build
+						    -processing: "No processor claimed any of these annotations" — the
+						                 animal-sniffer @IgnoreJRERequirement is consumed by the
+						                 maven-plugin (not a JSR-269 processor), so javac flags
+						                 it every build
+						  -Werror is on: the ElementType.MODULE blocker is gone now that
+						  @NullMarked lives only in module-info.java (compiled at release 9 in
+						  a separate execution); see module-info.java's javadoc.
+						-->
+						<arg>-Xlint:all,-serial,-options,-classfile,-processing</arg>
+						<arg>-Werror</arg>
+						<!--
+						  Checker Framework Nullness Checker runs as a second nullness pass
+						  alongside NullAway. CF is generics-aware and stricter than NullAway,
+						  so it acts as a second-opinion verifier on the same JSpecify
+						  annotations.
+						-->
+						<!-- Explicit processor list: Lombok (two SPI classes) + Checker Framework's
+							 Nullness Checker as a 2nd nullness pass alongside NullAway. The explicit
+							 -processor argument overrides SPI discovery, so every processor we want
+							 to run must be listed here. -->
+						<arg>-processor</arg>
+						<arg>lombok.launch.AnnotationProcessorHider$AnnotationProcessor,lombok.launch.AnnotationProcessorHider$ClaimingProcessor,org.checkerframework.checker.nullness.NullnessChecker</arg>
+						<arg>-XDaddTypeAnnotationsToSymbol=true</arg>
+						<arg>-XDcompilePolicy=simple</arg>
+						<arg>--should-stop=ifError=FLOW</arg>
+						<arg>-Xplugin:ErrorProne -Xep:NullAway:ERROR -XepOpt:NullAway:OnlyNullMarked=true -XepOpt:NullAway:JSpecifyMode=true -XepOpt:NullAway:CheckOptionalEmptiness=true -XepOpt:NullAway:AcknowledgeRestrictiveAnnotations=true -XepOpt:NullAway:AcknowledgeAndroidRecent=true -XepOpt:NullAway:AssertsEnabled=true -Xep:BoxedPrimitiveEquality:ERROR -Xep:EqualsHashCode:ERROR -Xep:EqualsIncompatibleType:ERROR -Xep:IdentityBinaryExpression:ERROR -Xep:SelfAssignment:ERROR -Xep:SelfComparison:ERROR -Xep:SelfEquals:ERROR -Xep:DeadException:ERROR -Xep:FormatString:ERROR -Xep:InvalidPatternSyntax:ERROR -Xep:OptionalEquality:ERROR -Xep:ImpossibleNullComparison:ERROR</arg>
+					</compilerArgs>
+					<annotationProcessorPaths>
+						<path>
+							<groupId>org.projectlombok</groupId>
+							<artifactId>lombok</artifactId>
+							<version>${lombok.version}</version>
+						</path>
+						<path>
+							<groupId>com.google.errorprone</groupId>
+							<artifactId>error_prone_core</artifactId>
+							<version>${errorprone.version}</version>
+						</path>
+						<path>
+							<groupId>com.uber.nullaway</groupId>
+							<artifactId>nullaway</artifactId>
+							<version>${nullaway.version}</version>
+						</path>
+						<path>
+							<groupId>org.checkerframework</groupId>
+							<artifactId>checker</artifactId>
+							<version>${checker.version}</version>
+						</path>
+					</annotationProcessorPaths>
+				</configuration>
+				<executions>
+					<execution>
+						<id>default-compile</id>
+						<configuration>
+							<!-- The default compile pass targets Java 8 bytecode for the library
+							     sources. module-info.java cannot be compiled at release 8 (it is a
+							     Java 9+ construct) and is handled by the separate execution below. -->
+							<excludes>
+								<exclude>module-info.java</exclude>
+							</excludes>
+						</configuration>
+					</execution>
+					<execution>
+						<id>module-info-compile</id>
+						<phase>compile</phase>
+						<goals>
+							<goal>compile</goal>
+						</goals>
+						<configuration>
+							<!-- module-info.java is the only Java 9+ source in this project; compile
+							     it at release 9 (the minimum version that understands modules) without
+							     the Error Prone / NullAway / Checker Framework processors that target
+							     ordinary source files only. Java 8 runtimes silently ignore the
+							     resulting module-info.class at the JAR root, so existing classpath
+							     consumers continue to work unchanged. LlamaLoader loads native libs
+							     via Class.getResourceAsStream on its own class object, which resolves
+							     to this module and therefore does not require any opens directives. -->
+							<release>9</release>
+							<includes>
+								<include>module-info.java</include>
+							</includes>
+							<compilerArgs combine.self="override"/>
+							<annotationProcessorPaths combine.self="override"/>
+						</configuration>
+					</execution>
+					<execution>
+						<id>default-testCompile</id>
+						<configuration>
+							<!-- Drop -parameters for test compile so jcstress reflection doesn't
+							     pick up parameter names from MethodParameters. -->
+							<parameters>false</parameters>
+							<compilerArgs combine.self="override">
+								<arg>-XDaddTypeAnnotationsToSymbol=true</arg>
+								<arg>-XDcompilePolicy=simple</arg>
+								<arg>--should-stop=ifError=FLOW</arg>
+								<arg>-Xplugin:ErrorProne -Xep:NullAway:OFF -Xep:GuardedBy:OFF</arg>
+							</compilerArgs>
+							<annotationProcessorPaths combine.children="append">
+								<path>
+									<groupId>org.openjdk.jcstress</groupId>
+									<artifactId>jcstress-core</artifactId>
+									<version>${jcstress.version}</version>
+								</path>
+								<path>
+									<groupId>org.openjdk.jmh</groupId>
+									<artifactId>jmh-generator-annprocess</artifactId>
+									<version>${jmh.version}</version>
+								</path>
+							</annotationProcessorPaths>
+						</configuration>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<artifactId>maven-resources-plugin</artifactId>
+			</plugin>
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-source-plugin</artifactId>
+				<executions>
+					<execution>
+						<id>attach-sources</id>
+						<goals>
+							<goal>jar-no-fork</goal>
+						</goals>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-javadoc-plugin</artifactId>
+				<configuration>
+					<source>1.8</source>
+					<failOnError>true</failOnError>
+					<failOnWarnings>true</failOnWarnings>
+					<doclint>all</doclint>
+				</configuration>
+				<executions>
+					<execution>
+						<id>attach-javadocs</id>
+						<goals>
+							<goal>jar</goal>
+						</goals>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<groupId>org.jacoco</groupId>
+				<artifactId>jacoco-maven-plugin</artifactId>
+				<executions>
+					<execution>
+						<id>prepare-agent</id>
+						<goals>
+							<goal>prepare-agent</goal>
+						</goals>
+					</execution>
+					<execution>
+						<id>report</id>
+						<phase>test</phase>
+						<goals>
+							<goal>report</goal>
+						</goals>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<!--
+				  Surefire is configured here only to control native-crash diagnostics:
+				  - argLine inherits jacoco's @{argLine} (so the jacoco runtime agent stays
+				    attached) and appends -XX:ErrorFile so the JVM crash log (hs_err_pid<pid>.log)
+				    always lands in the project basedir, which is the same directory the CI
+				    upload-artifact globs (${{ github.workspace }}/hs_err_pid*.log).
+				  - HeapDumpOnOutOfMemoryError + HeapDumpPath catch OOMs that would otherwise
+				    just print a stack and exit.
+				  - The version is pinned to the same release Maven would otherwise auto-select
+				    (visible in CI error logs as 3.5.4), so we don't drift independently of
+				    Maven's plugin-defaults bump.
+				-->
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-surefire-plugin</artifactId>
+				<configuration>
+					<argLine>@{argLine} -Xmx2g -XX:ErrorFile=hs_err_pid%p.log -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=.</argLine>
+					<!--
+					  Capture each test class's stdout/stderr into
+					  target/surefire-reports/<class>-output.txt. When a native crash
+					  takes down the forked JVM, this file is the ONLY record of
+					  which @Test method was last running (without it the surefire
+					  *.txt summary just says "VM terminated", which we already saw
+					  in CI without telling us which test was on fire).
+					-->
+					<redirectTestOutputToFile>true</redirectTestOutputToFile>
+				</configuration>
+			</plugin>
+			<plugin>
+				<groupId>com.diffplug.spotless</groupId>
+				<artifactId>spotless-maven-plugin</artifactId>
+				<configuration>
+					<java>
+						<includes>
+							<include>src/main/java/**/*.java</include>
+							<include>src/test/java/**/*.java</include>
+						</includes>
+						<palantirJavaFormat>
+							<version>${palantir-java-format.version}</version>
+						</palantirJavaFormat>
+						<removeUnusedImports/>
+						<trimTrailingWhitespace/>
+						<endWithNewline/>
+					</java>
+				</configuration>
+				<executions>
+					<execution>
+						<id>spotless-check</id>
+						<phase>verify</phase>
+						<goals>
+							<goal>check</goal>
+						</goals>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<groupId>com.github.spotbugs</groupId>
+				<artifactId>spotbugs-maven-plugin</artifactId>
+				<configuration>
+					<effort>Max</effort>
+					<threshold>Low</threshold>
+					<failOnError>true</failOnError>
+					<includeTests>false</includeTests>
+					<excludeFilterFile>spotbugs-exclude.xml</excludeFilterFile>
+					<plugins>
+						<plugin>
+							<groupId>com.mebigfatguy.fb-contrib</groupId>
+							<artifactId>fb-contrib</artifactId>
+							<version>${fb-contrib.version}</version>
+						</plugin>
+						<plugin>
+							<groupId>com.h3xstream.findsecbugs</groupId>
+							<artifactId>findsecbugs-plugin</artifactId>
+							<version>${findsecbugs.version}</version>
+						</plugin>
+					</plugins>
+				</configuration>
+				<executions>
+					<execution>
+						<id>spotbugs-check</id>
+						<phase>verify</phase>
+						<goals>
+							<goal>check</goal>
+						</goals>
+					</execution>
+				</executions>
+			</plugin>
+			<plugin>
+				<groupId>org.codehaus.mojo</groupId>
+				<artifactId>exec-maven-plugin</artifactId>
+				<configuration>
+					<mainClass>org.openjdk.jmh.Main</mainClass>
+					<classpathScope>test</classpathScope>
+				</configuration>
+			</plugin>
+			<plugin>
+				<!--
+				  Generic PIT policy (version, 100% gate, invocation, expansion rule):
+				  see ../workspace/policies/pit-mutation-testing.md.
+				  Repo-specific: the gated value/exception/args/json classes are pure-Java
+				  (no native libjllama or model file needed); targetClasses/targetTests are
+				  kept in lock-step. NOTE: value.ContentPart.audioFile(Path) mutants are only
+				  killed with the audio fixture present — see TODO.md and policy §4.
+				-->
+				<groupId>org.pitest</groupId>
+				<artifactId>pitest-maven</artifactId>
+				<dependencies>
+					<dependency>
+						<groupId>org.pitest</groupId>
+						<artifactId>pitest-junit5-plugin</artifactId>
+						<version>1.2.3</version>
+					</dependency>
+				</dependencies>
+				<configuration>
+					<targetClasses>
+						<param>net.ladenthin.llama.value.*</param>
+						<param>net.ladenthin.llama.exception.*</param>
+						<param>net.ladenthin.llama.args.*</param>
+						<param>net.ladenthin.llama.json.TimingsLogger</param>
+						<param>net.ladenthin.llama.json.RerankResponseParser</param>
+						<param>net.ladenthin.llama.json.ChatResponseParser</param>
+						<param>net.ladenthin.llama.json.CompletionResponseParser</param>
+					</targetClasses>
+					<targetTests>
+						<param>net.ladenthin.llama.value.*</param>
+						<param>net.ladenthin.llama.exception.*</param>
+						<param>net.ladenthin.llama.args.*</param>
+						<param>net.ladenthin.llama.json.*</param>
+					</targetTests>
+					<mutationThreshold>100</mutationThreshold>
+					<timeoutConstant>30000</timeoutConstant>
+				</configuration>
+			</plugin>
+		</plugins>
+	</build>
+
+	<profiles>
+		<!--
+			The `release` profile (GPG signing + Central Publishing) now lives in the parent
+			pom (../pom.xml) and is inherited, so a single reactor `mvn -P release deploy`
+			signs and publishes every module. The per-classifier native profiles below stay
+			here — they are core-specific.
+		-->
+		<profile>
+			<id>cuda</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-compiler-plugin</artifactId>
+						<executions>
+						<!-- We have to perform a separate build pass for cuda
+						classifier -->
+						<execution>
+							<id>gpu</id>
+							<phase>compile</phase>
+							<goals>
+								<goal>compile</goal>
+							</goals>
+							<configuration>
+								<!-- Same rationale as default-compile: this pass targets release 8;
+								     module-info.java is Java 9+ and is handled by module-info-compile. -->
+								<excludes>
+									<exclude>module-info.java</exclude>
+								</excludes>
+								<compilerArgs>
+									<arg>-h</arg>
+									<arg>src/main/cpp</arg>
+								</compilerArgs>
+								<outputDirectory>
+									${project.build.outputDirectory}_cuda</outputDirectory>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<artifactId>maven-resources-plugin</artifactId>
+						<executions>
+						<!-- Copy custom cuda libs to the output directory -->
+						<execution>
+							<id>copy-resources</id>
+							<phase>process-classes</phase>
+							<goals>
+								<goal>copy-resources</goal>
+							</goals>
+							<configuration>
+								<outputDirectory>
+									${project.build.outputDirectory}_cuda</outputDirectory>
+								<resources>
+									<resource>
+										<directory>
+											${basedir}/src/main/resources_linux_cuda/</directory>
+										<includes>
+											<include>**/*.*</include>
+										</includes>
+									</resource>
+								</resources>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-jar-plugin</artifactId>
+						<executions>
+							<!-- Pick class files AND libs from custom output
+							directory -->
+							<execution>
+								<id>cuda</id>
+								<phase>package</phase>
+								<goals>
+									<goal>jar</goal>
+								</goals>
+								<configuration>
+									<classifier>cuda13-linux-x86-64</classifier>
+									<classesDirectory>
+										${project.build.outputDirectory}_cuda</classesDirectory>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+
+		<profile>
+			<id>opencl-android</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-compiler-plugin</artifactId>
+						<executions>
+						<!-- Separate build pass for opencl-android classifier -->
+						<execution>
+							<id>opencl-android</id>
+							<phase>compile</phase>
+							<goals>
+								<goal>compile</goal>
+							</goals>
+							<configuration>
+								<!-- Same rationale as default-compile: this pass targets release 8;
+								     module-info.java is Java 9+ and is handled by module-info-compile. -->
+								<excludes>
+									<exclude>module-info.java</exclude>
+								</excludes>
+								<compilerArgs>
+									<arg>-h</arg>
+									<arg>src/main/cpp</arg>
+								</compilerArgs>
+								<outputDirectory>
+									${project.build.outputDirectory}_opencl_android</outputDirectory>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<artifactId>maven-resources-plugin</artifactId>
+						<executions>
+						<!-- Copy custom opencl-android libs to the output directory -->
+						<execution>
+							<id>copy-resources-opencl-android</id>
+							<phase>process-classes</phase>
+							<goals>
+								<goal>copy-resources</goal>
+							</goals>
+							<configuration>
+								<outputDirectory>
+									${project.build.outputDirectory}_opencl_android</outputDirectory>
+								<resources>
+									<resource>
+										<directory>
+											${basedir}/src/main/resources_android_opencl/</directory>
+										<includes>
+											<include>**/*.*</include>
+										</includes>
+									</resource>
+								</resources>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-jar-plugin</artifactId>
+						<executions>
+							<!-- Pick class files AND libs from custom output
+							directory -->
+							<execution>
+								<id>opencl-android</id>
+								<phase>package</phase>
+								<goals>
+									<goal>jar</goal>
+								</goals>
+								<configuration>
+									<classifier>opencl-android-aarch64</classifier>
+									<classesDirectory>
+										${project.build.outputDirectory}_opencl_android</classesDirectory>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+		<profile>
+			<!-- MSVC (Visual Studio generator) Windows CPU natives, shipped as the
+			     `msvc-windows` classifier JAR. After the default flip the DEFAULT JAR
+			     ships the Ninja Multi-Config natives (sccache-cached, merged into
+			     src/main/resources/ by CI); this profile packages the alternative
+			     MSVC-generator build for anyone who wants it. Both use the same MSVC
+			     toolchain (static /MT CRT) so they are functionally equivalent. Mirrors
+			     the cuda / opencl-android classifier pattern: a separate compile pass +
+			     resource copy + classified jar. The MSVC DLLs are staged into
+			     src/main/resources_windows_msvc/ by CI before this profile runs. -->
+			<id>windows-msvc</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-compiler-plugin</artifactId>
+						<executions>
+						<!-- Separate build pass for the windows-msvc classifier -->
+						<execution>
+							<id>windows-msvc</id>
+							<phase>compile</phase>
+							<goals>
+								<goal>compile</goal>
+							</goals>
+							<configuration>
+								<!-- Same rationale as default-compile: this pass targets release 8;
+								     module-info.java is Java 9+ and is handled by module-info-compile. -->
+								<excludes>
+									<exclude>module-info.java</exclude>
+								</excludes>
+								<compilerArgs>
+									<arg>-h</arg>
+									<arg>src/main/cpp</arg>
+								</compilerArgs>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_msvc</outputDirectory>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<artifactId>maven-resources-plugin</artifactId>
+						<executions>
+						<!-- Copy the Ninja-built Windows libs to the output directory -->
+						<execution>
+							<id>copy-resources-windows-msvc</id>
+							<phase>process-classes</phase>
+							<goals>
+								<goal>copy-resources</goal>
+							</goals>
+							<configuration>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_msvc</outputDirectory>
+								<resources>
+									<resource>
+										<directory>
+											${basedir}/src/main/resources_windows_msvc/</directory>
+										<includes>
+											<include>**/*.*</include>
+										</includes>
+									</resource>
+								</resources>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-jar-plugin</artifactId>
+						<executions>
+							<!-- Pick class files AND libs from custom output
+							directory -->
+							<execution>
+								<id>windows-msvc</id>
+								<phase>package</phase>
+								<goals>
+									<goal>jar</goal>
+								</goals>
+								<configuration>
+									<classifier>msvc-windows</classifier>
+									<classesDirectory>
+										${project.build.outputDirectory}_windows_msvc</classesDirectory>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+
+		<profile>
+			<!-- Windows CUDA GPU natives (x86_64 only), shipped as the
+			     `cuda13-windows-x86-64` classifier JAR. Built with -DGGML_CUDA=ON and
+			     routed by CMakeLists.txt to src/main/resources_windows_cuda/. The CUDA
+			     runtime libs (cudart64_13 / cublas64_13 / cublasLt64_13) are NOT bundled
+			     — the consumer's installed CUDA Toolkit provides them. Mirrors the
+			     cuda (Linux) classifier pattern. Staged by CI before this profile runs. -->
+			<id>cuda-windows</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-compiler-plugin</artifactId>
+						<executions>
+						<execution>
+							<id>cuda-windows</id>
+							<phase>compile</phase>
+							<goals>
+								<goal>compile</goal>
+							</goals>
+							<configuration>
+								<excludes>
+									<exclude>module-info.java</exclude>
+								</excludes>
+								<compilerArgs>
+									<arg>-h</arg>
+									<arg>src/main/cpp</arg>
+								</compilerArgs>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_cuda</outputDirectory>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<artifactId>maven-resources-plugin</artifactId>
+						<executions>
+						<execution>
+							<id>copy-resources-cuda-windows</id>
+							<phase>process-classes</phase>
+							<goals>
+								<goal>copy-resources</goal>
+							</goals>
+							<configuration>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_cuda</outputDirectory>
+								<resources>
+									<resource>
+										<directory>
+											${basedir}/src/main/resources_windows_cuda/</directory>
+										<includes>
+											<include>**/*.*</include>
+										</includes>
+									</resource>
+								</resources>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-jar-plugin</artifactId>
+						<executions>
+							<execution>
+								<id>cuda-windows</id>
+								<phase>package</phase>
+								<goals>
+									<goal>jar</goal>
+								</goals>
+								<configuration>
+									<classifier>cuda13-windows-x86-64</classifier>
+									<classesDirectory>
+										${project.build.outputDirectory}_windows_cuda</classesDirectory>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+
+		<profile>
+			<!-- Windows Vulkan GPU natives (x86_64 only), shipped as the
+			     `vulkan-windows-x86-64` classifier JAR. Built with -DGGML_VULKAN=ON and
+			     routed by CMakeLists.txt to src/main/resources_windows_vulkan/. The Vulkan
+			     runtime (vulkan-1.dll) ships with the GPU driver — nothing is bundled.
+			     Vulkan is the most portable Windows GPU backend (NVIDIA/AMD/Intel).
+			     Staged by CI before this profile runs. -->
+			<id>vulkan-windows</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-compiler-plugin</artifactId>
+						<executions>
+						<execution>
+							<id>vulkan-windows</id>
+							<phase>compile</phase>
+							<goals>
+								<goal>compile</goal>
+							</goals>
+							<configuration>
+								<excludes>
+									<exclude>module-info.java</exclude>
+								</excludes>
+								<compilerArgs>
+									<arg>-h</arg>
+									<arg>src/main/cpp</arg>
+								</compilerArgs>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_vulkan</outputDirectory>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<artifactId>maven-resources-plugin</artifactId>
+						<executions>
+						<execution>
+							<id>copy-resources-vulkan-windows</id>
+							<phase>process-classes</phase>
+							<goals>
+								<goal>copy-resources</goal>
+							</goals>
+							<configuration>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_vulkan</outputDirectory>
+								<resources>
+									<resource>
+										<directory>
+											${basedir}/src/main/resources_windows_vulkan/</directory>
+										<includes>
+											<include>**/*.*</include>
+										</includes>
+									</resource>
+								</resources>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-jar-plugin</artifactId>
+						<executions>
+							<execution>
+								<id>vulkan-windows</id>
+								<phase>package</phase>
+								<goals>
+									<goal>jar</goal>
+								</goals>
+								<configuration>
+									<classifier>vulkan-windows-x86-64</classifier>
+									<classesDirectory>
+										${project.build.outputDirectory}_windows_vulkan</classesDirectory>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+
+		<profile>
+			<!-- Windows OpenCL GPU natives (x86_64 only), shipped as the
+			     `opencl-windows-x86-64` classifier JAR. Built with -DGGML_OPENCL=ON and
+			     routed by CMakeLists.txt to src/main/resources_windows_opencl/. The OpenCL
+			     ICD (System32\OpenCL.dll) ships with the GPU driver — nothing is bundled.
+			     NOTE: the GGML OpenCL backend is Adreno-tuned; on desktop GPUs CUDA/Vulkan
+			     are better supported. Staged by CI before this profile runs. -->
+			<id>opencl-windows</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-compiler-plugin</artifactId>
+						<executions>
+						<execution>
+							<id>opencl-windows</id>
+							<phase>compile</phase>
+							<goals>
+								<goal>compile</goal>
+							</goals>
+							<configuration>
+								<excludes>
+									<exclude>module-info.java</exclude>
+								</excludes>
+								<compilerArgs>
+									<arg>-h</arg>
+									<arg>src/main/cpp</arg>
+								</compilerArgs>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_opencl</outputDirectory>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<artifactId>maven-resources-plugin</artifactId>
+						<executions>
+						<execution>
+							<id>copy-resources-opencl-windows</id>
+							<phase>process-classes</phase>
+							<goals>
+								<goal>copy-resources</goal>
+							</goals>
+							<configuration>
+								<outputDirectory>
+									${project.build.outputDirectory}_windows_opencl</outputDirectory>
+								<resources>
+									<resource>
+										<directory>
+											${basedir}/src/main/resources_windows_opencl/</directory>
+										<includes>
+											<include>**/*.*</include>
+										</includes>
+									</resource>
+								</resources>
+							</configuration>
+						</execution>
+						</executions>
+					</plugin>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-jar-plugin</artifactId>
+						<executions>
+							<execution>
+								<id>opencl-windows</id>
+								<phase>package</phase>
+								<goals>
+									<goal>jar</goal>
+								</goals>
+								<configuration>
+									<classifier>opencl-windows-x86-64</classifier>
+									<classesDirectory>
+										${project.build.outputDirectory}_windows_opencl</classesDirectory>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+
+		<profile>
+			<id>vmlens</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>com.vmlens</groupId>
+						<artifactId>vmlens-maven-plugin</artifactId>
+						<configuration>
+							<!--
+								Run vmlens interleaving analysis over the whole `vmlens` test
+								package (smoke test + SessionStateInterleavingTest). The `com.vmlens:api`
+								test dependency lives in the main <dependencies> block.
+								Expand <includes> as more concurrency tests are added (the
+								streambuffer repo runs vmlens over its whole suite).
+							-->
+							<includes>
+								<include>**/vmlens/*.java</include>
+							</includes>
+						</configuration>
+						<executions>
+							<execution>
+								<id>vmlens-test</id>
+								<goals>
+									<goal>test</goal>
+								</goals>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+		<profile>
+			<id>jcstress</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.codehaus.mojo</groupId>
+						<artifactId>exec-maven-plugin</artifactId>
+						<executions>
+							<execution>
+								<id>jcstress</id>
+								<phase>test</phase>
+								<goals><goal>exec</goal></goals>
+								<configuration>
+									<executable>${java.home}/bin/java</executable>
+									<classpathScope>test</classpathScope>
+									<arguments>
+										<argument>-classpath</argument>
+										<classpath/>
+										<argument>org.openjdk.jcstress.Main</argument>
+										<argument>-v</argument>
+										<argument>-m</argument>
+										<argument>default</argument>
+									</arguments>
+								</configuration>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+		<profile>
+			<!--
+				Builds the fat jar-with-dependencies uber JAR: the library classes, the
+				default-platform native libs from src/main/resources, and all runtime Java
+				dependencies in one drop-on-classpath JAR, runnable via the OpenAiCompatServer
+				Main-Class (set below) to start the OpenAI-compatible HTTP server. Off by
+				default; the CI `package` job activates it so the uber JAR rides along in the
+				`llama-jars` upload-artifact bundle. Documented in CLAUDE.md "Build Commands"
+				as `mvn -P assembly package`.
+			-->
+			<id>assembly</id>
+			<build>
+				<plugins>
+					<plugin>
+						<groupId>org.apache.maven.plugins</groupId>
+						<artifactId>maven-assembly-plugin</artifactId>
+						<configuration>
+							<descriptorRefs>
+								<descriptorRef>jar-with-dependencies</descriptorRef>
+							</descriptorRefs>
+							<archive>
+								<manifest>
+									<mainClass>net.ladenthin.llama.server.OpenAiCompatServer</mainClass>
+								</manifest>
+							</archive>
+						</configuration>
+						<executions>
+							<execution>
+								<id>build-fat-jar</id>
+								<phase>package</phase>
+								<goals>
+									<goal>single</goal>
+								</goals>
+							</execution>
+						</executions>
+					</plugin>
+				</plugins>
+			</build>
+		</profile>
+	</profiles>
+</project>
diff --git a/spotbugs-exclude.xml b/llama/spotbugs-exclude.xml
similarity index 100%
rename from spotbugs-exclude.xml
rename to llama/spotbugs-exclude.xml
diff --git a/src/main/cpp/compat/ggml_x86_compat.c b/llama/src/main/cpp/compat/ggml_x86_compat.c
similarity index 100%
rename from src/main/cpp/compat/ggml_x86_compat.c
rename to llama/src/main/cpp/compat/ggml_x86_compat.c
diff --git a/src/main/cpp/jllama.cpp b/llama/src/main/cpp/jllama.cpp
similarity index 99%
rename from src/main/cpp/jllama.cpp
rename to llama/src/main/cpp/jllama.cpp
index c0eef843..f3871f89 100644
--- a/src/main/cpp/jllama.cpp
+++ b/llama/src/main/cpp/jllama.cpp
@@ -770,9 +770,8 @@ JNIEXPORT void JNICALL Java_net_ladenthin_llama_LlamaModel_loadModelWithProgress
 // Build the special-token id map (a token is -1 / LLAMA_TOKEN_NULL when the model defines none).
 static json special_tokens_json(const llama_vocab *vocab) {
     return {
-        {"bos", llama_vocab_bos(vocab)}, {"eos", llama_vocab_eos(vocab)},
-        {"eot", llama_vocab_eot(vocab)}, {"sep", llama_vocab_sep(vocab)},
-        {"nl", llama_vocab_nl(vocab)},   {"pad", llama_vocab_pad(vocab)},
+        {"bos", llama_vocab_bos(vocab)}, {"eos", llama_vocab_eos(vocab)}, {"eot", llama_vocab_eot(vocab)},
+        {"sep", llama_vocab_sep(vocab)}, {"nl", llama_vocab_nl(vocab)},   {"pad", llama_vocab_pad(vocab)},
     };
 }
 
diff --git a/src/main/cpp/jllama.h b/llama/src/main/cpp/jllama.h
similarity index 100%
rename from src/main/cpp/jllama.h
rename to llama/src/main/cpp/jllama.h
diff --git a/src/main/cpp/jni_helpers.hpp b/llama/src/main/cpp/jni_helpers.hpp
similarity index 100%
rename from src/main/cpp/jni_helpers.hpp
rename to llama/src/main/cpp/jni_helpers.hpp
diff --git a/src/main/cpp/json_helpers.hpp b/llama/src/main/cpp/json_helpers.hpp
similarity index 100%
rename from src/main/cpp/json_helpers.hpp
rename to llama/src/main/cpp/json_helpers.hpp
diff --git a/src/main/cpp/log_helpers.hpp b/llama/src/main/cpp/log_helpers.hpp
similarity index 100%
rename from src/main/cpp/log_helpers.hpp
rename to llama/src/main/cpp/log_helpers.hpp
diff --git a/src/main/cpp/tts_engine.cpp b/llama/src/main/cpp/tts_engine.cpp
similarity index 100%
rename from src/main/cpp/tts_engine.cpp
rename to llama/src/main/cpp/tts_engine.cpp
diff --git a/src/main/cpp/tts_engine.h b/llama/src/main/cpp/tts_engine.h
similarity index 100%
rename from src/main/cpp/tts_engine.h
rename to llama/src/main/cpp/tts_engine.h
diff --git a/src/main/cpp/tts_upstream.h b/llama/src/main/cpp/tts_upstream.h
similarity index 100%
rename from src/main/cpp/tts_upstream.h
rename to llama/src/main/cpp/tts_upstream.h
diff --git a/src/main/cpp/tts_wav.hpp b/llama/src/main/cpp/tts_wav.hpp
similarity index 100%
rename from src/main/cpp/tts_wav.hpp
rename to llama/src/main/cpp/tts_wav.hpp
diff --git a/src/main/cpp/utils.hpp b/llama/src/main/cpp/utils.hpp
similarity index 100%
rename from src/main/cpp/utils.hpp
rename to llama/src/main/cpp/utils.hpp
diff --git a/src/main/cpp/webui_stub/ui.h b/llama/src/main/cpp/webui_stub/ui.h
similarity index 100%
rename from src/main/cpp/webui_stub/ui.h
rename to llama/src/main/cpp/webui_stub/ui.h
diff --git a/src/main/java/module-info.java b/llama/src/main/java/module-info.java
similarity index 100%
rename from src/main/java/module-info.java
rename to llama/src/main/java/module-info.java
diff --git a/src/main/java/net/ladenthin/llama/LlamaIterable.java b/llama/src/main/java/net/ladenthin/llama/LlamaIterable.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/LlamaIterable.java
rename to llama/src/main/java/net/ladenthin/llama/LlamaIterable.java
diff --git a/src/main/java/net/ladenthin/llama/LlamaIterator.java b/llama/src/main/java/net/ladenthin/llama/LlamaIterator.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/LlamaIterator.java
rename to llama/src/main/java/net/ladenthin/llama/LlamaIterator.java
diff --git a/src/main/java/net/ladenthin/llama/LlamaModel.java b/llama/src/main/java/net/ladenthin/llama/LlamaModel.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/LlamaModel.java
rename to llama/src/main/java/net/ladenthin/llama/LlamaModel.java
diff --git a/src/main/java/net/ladenthin/llama/Session.java b/llama/src/main/java/net/ladenthin/llama/Session.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/Session.java
rename to llama/src/main/java/net/ladenthin/llama/Session.java
diff --git a/src/main/java/net/ladenthin/llama/SessionState.java b/llama/src/main/java/net/ladenthin/llama/SessionState.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/SessionState.java
rename to llama/src/main/java/net/ladenthin/llama/SessionState.java
diff --git a/src/main/java/net/ladenthin/llama/TextToSpeech.java b/llama/src/main/java/net/ladenthin/llama/TextToSpeech.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/TextToSpeech.java
rename to llama/src/main/java/net/ladenthin/llama/TextToSpeech.java
diff --git a/src/main/java/net/ladenthin/llama/ToolCallingAgent.java b/llama/src/main/java/net/ladenthin/llama/ToolCallingAgent.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/ToolCallingAgent.java
rename to llama/src/main/java/net/ladenthin/llama/ToolCallingAgent.java
diff --git a/src/main/java/net/ladenthin/llama/args/CacheType.java b/llama/src/main/java/net/ladenthin/llama/args/CacheType.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/CacheType.java
rename to llama/src/main/java/net/ladenthin/llama/args/CacheType.java
diff --git a/src/main/java/net/ladenthin/llama/args/CliArg.java b/llama/src/main/java/net/ladenthin/llama/args/CliArg.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/CliArg.java
rename to llama/src/main/java/net/ladenthin/llama/args/CliArg.java
diff --git a/src/main/java/net/ladenthin/llama/args/ContinuationMode.java b/llama/src/main/java/net/ladenthin/llama/args/ContinuationMode.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/ContinuationMode.java
rename to llama/src/main/java/net/ladenthin/llama/args/ContinuationMode.java
diff --git a/src/main/java/net/ladenthin/llama/args/GpuSplitMode.java b/llama/src/main/java/net/ladenthin/llama/args/GpuSplitMode.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/GpuSplitMode.java
rename to llama/src/main/java/net/ladenthin/llama/args/GpuSplitMode.java
diff --git a/src/main/java/net/ladenthin/llama/args/LogFormat.java b/llama/src/main/java/net/ladenthin/llama/args/LogFormat.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/LogFormat.java
rename to llama/src/main/java/net/ladenthin/llama/args/LogFormat.java
diff --git a/src/main/java/net/ladenthin/llama/args/MiroStat.java b/llama/src/main/java/net/ladenthin/llama/args/MiroStat.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/MiroStat.java
rename to llama/src/main/java/net/ladenthin/llama/args/MiroStat.java
diff --git a/src/main/java/net/ladenthin/llama/args/ModelFlag.java b/llama/src/main/java/net/ladenthin/llama/args/ModelFlag.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/ModelFlag.java
rename to llama/src/main/java/net/ladenthin/llama/args/ModelFlag.java
diff --git a/src/main/java/net/ladenthin/llama/args/NumaStrategy.java b/llama/src/main/java/net/ladenthin/llama/args/NumaStrategy.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/NumaStrategy.java
rename to llama/src/main/java/net/ladenthin/llama/args/NumaStrategy.java
diff --git a/src/main/java/net/ladenthin/llama/args/PoolingType.java b/llama/src/main/java/net/ladenthin/llama/args/PoolingType.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/PoolingType.java
rename to llama/src/main/java/net/ladenthin/llama/args/PoolingType.java
diff --git a/src/main/java/net/ladenthin/llama/args/ReasoningFormat.java b/llama/src/main/java/net/ladenthin/llama/args/ReasoningFormat.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/ReasoningFormat.java
rename to llama/src/main/java/net/ladenthin/llama/args/ReasoningFormat.java
diff --git a/src/main/java/net/ladenthin/llama/args/RopeScalingType.java b/llama/src/main/java/net/ladenthin/llama/args/RopeScalingType.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/RopeScalingType.java
rename to llama/src/main/java/net/ladenthin/llama/args/RopeScalingType.java
diff --git a/src/main/java/net/ladenthin/llama/args/Sampler.java b/llama/src/main/java/net/ladenthin/llama/args/Sampler.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/Sampler.java
rename to llama/src/main/java/net/ladenthin/llama/args/Sampler.java
diff --git a/src/main/java/net/ladenthin/llama/args/package-info.java b/llama/src/main/java/net/ladenthin/llama/args/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/args/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/args/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/callback/CancellationToken.java b/llama/src/main/java/net/ladenthin/llama/callback/CancellationToken.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/callback/CancellationToken.java
rename to llama/src/main/java/net/ladenthin/llama/callback/CancellationToken.java
diff --git a/src/main/java/net/ladenthin/llama/callback/LoadProgressCallback.java b/llama/src/main/java/net/ladenthin/llama/callback/LoadProgressCallback.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/callback/LoadProgressCallback.java
rename to llama/src/main/java/net/ladenthin/llama/callback/LoadProgressCallback.java
diff --git a/src/main/java/net/ladenthin/llama/callback/ToolHandler.java b/llama/src/main/java/net/ladenthin/llama/callback/ToolHandler.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/callback/ToolHandler.java
rename to llama/src/main/java/net/ladenthin/llama/callback/ToolHandler.java
diff --git a/src/main/java/net/ladenthin/llama/callback/package-info.java b/llama/src/main/java/net/ladenthin/llama/callback/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/callback/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/callback/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/exception/LlamaException.java b/llama/src/main/java/net/ladenthin/llama/exception/LlamaException.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/exception/LlamaException.java
rename to llama/src/main/java/net/ladenthin/llama/exception/LlamaException.java
diff --git a/src/main/java/net/ladenthin/llama/exception/ModelUnavailableException.java b/llama/src/main/java/net/ladenthin/llama/exception/ModelUnavailableException.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/exception/ModelUnavailableException.java
rename to llama/src/main/java/net/ladenthin/llama/exception/ModelUnavailableException.java
diff --git a/src/main/java/net/ladenthin/llama/exception/package-info.java b/llama/src/main/java/net/ladenthin/llama/exception/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/exception/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/exception/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/json/ChatResponseParser.java b/llama/src/main/java/net/ladenthin/llama/json/ChatResponseParser.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/json/ChatResponseParser.java
rename to llama/src/main/java/net/ladenthin/llama/json/ChatResponseParser.java
diff --git a/src/main/java/net/ladenthin/llama/json/ChatStreamChunkParser.java b/llama/src/main/java/net/ladenthin/llama/json/ChatStreamChunkParser.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/json/ChatStreamChunkParser.java
rename to llama/src/main/java/net/ladenthin/llama/json/ChatStreamChunkParser.java
diff --git a/src/main/java/net/ladenthin/llama/json/CompletionResponseParser.java b/llama/src/main/java/net/ladenthin/llama/json/CompletionResponseParser.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/json/CompletionResponseParser.java
rename to llama/src/main/java/net/ladenthin/llama/json/CompletionResponseParser.java
diff --git a/src/main/java/net/ladenthin/llama/json/RerankResponseParser.java b/llama/src/main/java/net/ladenthin/llama/json/RerankResponseParser.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/json/RerankResponseParser.java
rename to llama/src/main/java/net/ladenthin/llama/json/RerankResponseParser.java
diff --git a/src/main/java/net/ladenthin/llama/json/TimingsLogger.java b/llama/src/main/java/net/ladenthin/llama/json/TimingsLogger.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/json/TimingsLogger.java
rename to llama/src/main/java/net/ladenthin/llama/json/TimingsLogger.java
diff --git a/src/main/java/net/ladenthin/llama/json/package-info.java b/llama/src/main/java/net/ladenthin/llama/json/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/json/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/json/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/loader/Java8CompatibilityHelper.java b/llama/src/main/java/net/ladenthin/llama/loader/Java8CompatibilityHelper.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/Java8CompatibilityHelper.java
rename to llama/src/main/java/net/ladenthin/llama/loader/Java8CompatibilityHelper.java
diff --git a/src/main/java/net/ladenthin/llama/loader/LlamaLoader.java b/llama/src/main/java/net/ladenthin/llama/loader/LlamaLoader.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/LlamaLoader.java
rename to llama/src/main/java/net/ladenthin/llama/loader/LlamaLoader.java
diff --git a/src/main/java/net/ladenthin/llama/loader/LlamaSystemProperties.java b/llama/src/main/java/net/ladenthin/llama/loader/LlamaSystemProperties.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/LlamaSystemProperties.java
rename to llama/src/main/java/net/ladenthin/llama/loader/LlamaSystemProperties.java
diff --git a/src/main/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetter.java b/llama/src/main/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetter.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetter.java
rename to llama/src/main/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetter.java
diff --git a/src/main/java/net/ladenthin/llama/loader/OSInfo.java b/llama/src/main/java/net/ladenthin/llama/loader/OSInfo.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/OSInfo.java
rename to llama/src/main/java/net/ladenthin/llama/loader/OSInfo.java
diff --git a/src/main/java/net/ladenthin/llama/loader/OfflineModelGuard.java b/llama/src/main/java/net/ladenthin/llama/loader/OfflineModelGuard.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/OfflineModelGuard.java
rename to llama/src/main/java/net/ladenthin/llama/loader/OfflineModelGuard.java
diff --git a/src/main/java/net/ladenthin/llama/loader/ProcessRunner.java b/llama/src/main/java/net/ladenthin/llama/loader/ProcessRunner.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/ProcessRunner.java
rename to llama/src/main/java/net/ladenthin/llama/loader/ProcessRunner.java
diff --git a/src/main/java/net/ladenthin/llama/loader/package-info.java b/llama/src/main/java/net/ladenthin/llama/loader/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/loader/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/loader/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/package-info.java b/llama/src/main/java/net/ladenthin/llama/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/ChatRequest.java b/llama/src/main/java/net/ladenthin/llama/parameters/ChatRequest.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/ChatRequest.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/ChatRequest.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/CliParameters.java b/llama/src/main/java/net/ladenthin/llama/parameters/CliParameters.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/CliParameters.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/CliParameters.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/InferenceParameters.java b/llama/src/main/java/net/ladenthin/llama/parameters/InferenceParameters.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/InferenceParameters.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/InferenceParameters.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/JsonParameters.java b/llama/src/main/java/net/ladenthin/llama/parameters/JsonParameters.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/JsonParameters.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/JsonParameters.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/ModelParameters.java b/llama/src/main/java/net/ladenthin/llama/parameters/ModelParameters.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/ModelParameters.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/ModelParameters.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/ParameterJsonSerializer.java b/llama/src/main/java/net/ladenthin/llama/parameters/ParameterJsonSerializer.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/ParameterJsonSerializer.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/ParameterJsonSerializer.java
diff --git a/src/main/java/net/ladenthin/llama/parameters/package-info.java b/llama/src/main/java/net/ladenthin/llama/parameters/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/parameters/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/parameters/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/server/AnthropicApiSupport.java b/llama/src/main/java/net/ladenthin/llama/server/AnthropicApiSupport.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/AnthropicApiSupport.java
rename to llama/src/main/java/net/ladenthin/llama/server/AnthropicApiSupport.java
diff --git a/src/main/java/net/ladenthin/llama/server/AnthropicStreamTranslator.java b/llama/src/main/java/net/ladenthin/llama/server/AnthropicStreamTranslator.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/AnthropicStreamTranslator.java
rename to llama/src/main/java/net/ladenthin/llama/server/AnthropicStreamTranslator.java
diff --git a/src/main/java/net/ladenthin/llama/server/ChunkSink.java b/llama/src/main/java/net/ladenthin/llama/server/ChunkSink.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/ChunkSink.java
rename to llama/src/main/java/net/ladenthin/llama/server/ChunkSink.java
diff --git a/src/main/java/net/ladenthin/llama/server/LlamaModelBackend.java b/llama/src/main/java/net/ladenthin/llama/server/LlamaModelBackend.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/LlamaModelBackend.java
rename to llama/src/main/java/net/ladenthin/llama/server/LlamaModelBackend.java
diff --git a/src/main/java/net/ladenthin/llama/server/NativeServer.java b/llama/src/main/java/net/ladenthin/llama/server/NativeServer.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/NativeServer.java
rename to llama/src/main/java/net/ladenthin/llama/server/NativeServer.java
diff --git a/src/main/java/net/ladenthin/llama/server/OaiRerankSupport.java b/llama/src/main/java/net/ladenthin/llama/server/OaiRerankSupport.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OaiRerankSupport.java
rename to llama/src/main/java/net/ladenthin/llama/server/OaiRerankSupport.java
diff --git a/src/main/java/net/ladenthin/llama/server/OllamaApiSupport.java b/llama/src/main/java/net/ladenthin/llama/server/OllamaApiSupport.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OllamaApiSupport.java
rename to llama/src/main/java/net/ladenthin/llama/server/OllamaApiSupport.java
diff --git a/src/main/java/net/ladenthin/llama/server/OpenAiBackend.java b/llama/src/main/java/net/ladenthin/llama/server/OpenAiBackend.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OpenAiBackend.java
rename to llama/src/main/java/net/ladenthin/llama/server/OpenAiBackend.java
diff --git a/src/main/java/net/ladenthin/llama/server/OpenAiCompatServer.java b/llama/src/main/java/net/ladenthin/llama/server/OpenAiCompatServer.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OpenAiCompatServer.java
rename to llama/src/main/java/net/ladenthin/llama/server/OpenAiCompatServer.java
diff --git a/src/main/java/net/ladenthin/llama/server/OpenAiRequestMapper.java b/llama/src/main/java/net/ladenthin/llama/server/OpenAiRequestMapper.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OpenAiRequestMapper.java
rename to llama/src/main/java/net/ladenthin/llama/server/OpenAiRequestMapper.java
diff --git a/src/main/java/net/ladenthin/llama/server/OpenAiServerCli.java b/llama/src/main/java/net/ladenthin/llama/server/OpenAiServerCli.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OpenAiServerCli.java
rename to llama/src/main/java/net/ladenthin/llama/server/OpenAiServerCli.java
diff --git a/src/main/java/net/ladenthin/llama/server/OpenAiServerConfig.java b/llama/src/main/java/net/ladenthin/llama/server/OpenAiServerConfig.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OpenAiServerConfig.java
rename to llama/src/main/java/net/ladenthin/llama/server/OpenAiServerConfig.java
diff --git a/src/main/java/net/ladenthin/llama/server/OpenAiSseFormatter.java b/llama/src/main/java/net/ladenthin/llama/server/OpenAiSseFormatter.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/OpenAiSseFormatter.java
rename to llama/src/main/java/net/ladenthin/llama/server/OpenAiSseFormatter.java
diff --git a/src/main/java/net/ladenthin/llama/server/ResponsesApiSupport.java b/llama/src/main/java/net/ladenthin/llama/server/ResponsesApiSupport.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/ResponsesApiSupport.java
rename to llama/src/main/java/net/ladenthin/llama/server/ResponsesApiSupport.java
diff --git a/src/main/java/net/ladenthin/llama/server/ResponsesStreamTranslator.java b/llama/src/main/java/net/ladenthin/llama/server/ResponsesStreamTranslator.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/ResponsesStreamTranslator.java
rename to llama/src/main/java/net/ladenthin/llama/server/ResponsesStreamTranslator.java
diff --git a/src/main/java/net/ladenthin/llama/server/ToolCallDeltaAccumulator.java b/llama/src/main/java/net/ladenthin/llama/server/ToolCallDeltaAccumulator.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/ToolCallDeltaAccumulator.java
rename to llama/src/main/java/net/ladenthin/llama/server/ToolCallDeltaAccumulator.java
diff --git a/src/main/java/net/ladenthin/llama/server/package-info.java b/llama/src/main/java/net/ladenthin/llama/server/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/server/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/server/package-info.java
diff --git a/src/main/java/net/ladenthin/llama/value/ChatChoice.java b/llama/src/main/java/net/ladenthin/llama/value/ChatChoice.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ChatChoice.java
rename to llama/src/main/java/net/ladenthin/llama/value/ChatChoice.java
diff --git a/src/main/java/net/ladenthin/llama/value/ChatMessage.java b/llama/src/main/java/net/ladenthin/llama/value/ChatMessage.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ChatMessage.java
rename to llama/src/main/java/net/ladenthin/llama/value/ChatMessage.java
diff --git a/src/main/java/net/ladenthin/llama/value/ChatResponse.java b/llama/src/main/java/net/ladenthin/llama/value/ChatResponse.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ChatResponse.java
rename to llama/src/main/java/net/ladenthin/llama/value/ChatResponse.java
diff --git a/src/main/java/net/ladenthin/llama/value/ChatTranscript.java b/llama/src/main/java/net/ladenthin/llama/value/ChatTranscript.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ChatTranscript.java
rename to llama/src/main/java/net/ladenthin/llama/value/ChatTranscript.java
diff --git a/src/main/java/net/ladenthin/llama/value/CompletionResult.java b/llama/src/main/java/net/ladenthin/llama/value/CompletionResult.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/CompletionResult.java
rename to llama/src/main/java/net/ladenthin/llama/value/CompletionResult.java
diff --git a/src/main/java/net/ladenthin/llama/value/ContentPart.java b/llama/src/main/java/net/ladenthin/llama/value/ContentPart.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ContentPart.java
rename to llama/src/main/java/net/ladenthin/llama/value/ContentPart.java
diff --git a/src/main/java/net/ladenthin/llama/value/LlamaOutput.java b/llama/src/main/java/net/ladenthin/llama/value/LlamaOutput.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/LlamaOutput.java
rename to llama/src/main/java/net/ladenthin/llama/value/LlamaOutput.java
diff --git a/src/main/java/net/ladenthin/llama/value/LogLevel.java b/llama/src/main/java/net/ladenthin/llama/value/LogLevel.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/LogLevel.java
rename to llama/src/main/java/net/ladenthin/llama/value/LogLevel.java
diff --git a/src/main/java/net/ladenthin/llama/value/ModelMeta.java b/llama/src/main/java/net/ladenthin/llama/value/ModelMeta.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ModelMeta.java
rename to llama/src/main/java/net/ladenthin/llama/value/ModelMeta.java
diff --git a/src/main/java/net/ladenthin/llama/value/Pair.java b/llama/src/main/java/net/ladenthin/llama/value/Pair.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/Pair.java
rename to llama/src/main/java/net/ladenthin/llama/value/Pair.java
diff --git a/src/main/java/net/ladenthin/llama/value/ServerMetrics.java b/llama/src/main/java/net/ladenthin/llama/value/ServerMetrics.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ServerMetrics.java
rename to llama/src/main/java/net/ladenthin/llama/value/ServerMetrics.java
diff --git a/src/main/java/net/ladenthin/llama/value/SlotMetrics.java b/llama/src/main/java/net/ladenthin/llama/value/SlotMetrics.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/SlotMetrics.java
rename to llama/src/main/java/net/ladenthin/llama/value/SlotMetrics.java
diff --git a/src/main/java/net/ladenthin/llama/value/StopReason.java b/llama/src/main/java/net/ladenthin/llama/value/StopReason.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/StopReason.java
rename to llama/src/main/java/net/ladenthin/llama/value/StopReason.java
diff --git a/src/main/java/net/ladenthin/llama/value/Timings.java b/llama/src/main/java/net/ladenthin/llama/value/Timings.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/Timings.java
rename to llama/src/main/java/net/ladenthin/llama/value/Timings.java
diff --git a/src/main/java/net/ladenthin/llama/value/TokenLogprob.java b/llama/src/main/java/net/ladenthin/llama/value/TokenLogprob.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/TokenLogprob.java
rename to llama/src/main/java/net/ladenthin/llama/value/TokenLogprob.java
diff --git a/src/main/java/net/ladenthin/llama/value/ToolCall.java b/llama/src/main/java/net/ladenthin/llama/value/ToolCall.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ToolCall.java
rename to llama/src/main/java/net/ladenthin/llama/value/ToolCall.java
diff --git a/src/main/java/net/ladenthin/llama/value/ToolDefinition.java b/llama/src/main/java/net/ladenthin/llama/value/ToolDefinition.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/ToolDefinition.java
rename to llama/src/main/java/net/ladenthin/llama/value/ToolDefinition.java
diff --git a/src/main/java/net/ladenthin/llama/value/Usage.java b/llama/src/main/java/net/ladenthin/llama/value/Usage.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/Usage.java
rename to llama/src/main/java/net/ladenthin/llama/value/Usage.java
diff --git a/src/main/java/net/ladenthin/llama/value/package-info.java b/llama/src/main/java/net/ladenthin/llama/value/package-info.java
similarity index 100%
rename from src/main/java/net/ladenthin/llama/value/package-info.java
rename to llama/src/main/java/net/ladenthin/llama/value/package-info.java
diff --git a/src/test/cpp/test_jni_helpers.cpp b/llama/src/test/cpp/test_jni_helpers.cpp
similarity index 100%
rename from src/test/cpp/test_jni_helpers.cpp
rename to llama/src/test/cpp/test_jni_helpers.cpp
diff --git a/src/test/cpp/test_json_helpers.cpp b/llama/src/test/cpp/test_json_helpers.cpp
similarity index 100%
rename from src/test/cpp/test_json_helpers.cpp
rename to llama/src/test/cpp/test_json_helpers.cpp
diff --git a/src/test/cpp/test_log_helpers.cpp b/llama/src/test/cpp/test_log_helpers.cpp
similarity index 100%
rename from src/test/cpp/test_log_helpers.cpp
rename to llama/src/test/cpp/test_log_helpers.cpp
diff --git a/src/test/cpp/test_server.cpp b/llama/src/test/cpp/test_server.cpp
similarity index 100%
rename from src/test/cpp/test_server.cpp
rename to llama/src/test/cpp/test_server.cpp
diff --git a/src/test/cpp/test_tts_wav.cpp b/llama/src/test/cpp/test_tts_wav.cpp
similarity index 100%
rename from src/test/cpp/test_tts_wav.cpp
rename to llama/src/test/cpp/test_tts_wav.cpp
diff --git a/src/test/cpp/test_utils.cpp b/llama/src/test/cpp/test_utils.cpp
similarity index 100%
rename from src/test/cpp/test_utils.cpp
rename to llama/src/test/cpp/test_utils.cpp
diff --git a/src/test/java/examples/ChatExample.java b/llama/src/test/java/examples/ChatExample.java
similarity index 100%
rename from src/test/java/examples/ChatExample.java
rename to llama/src/test/java/examples/ChatExample.java
diff --git a/src/test/java/examples/GrammarExample.java b/llama/src/test/java/examples/GrammarExample.java
similarity index 100%
rename from src/test/java/examples/GrammarExample.java
rename to llama/src/test/java/examples/GrammarExample.java
diff --git a/src/test/java/examples/InfillExample.java b/llama/src/test/java/examples/InfillExample.java
similarity index 100%
rename from src/test/java/examples/InfillExample.java
rename to llama/src/test/java/examples/InfillExample.java
diff --git a/src/test/java/examples/MainExample.java b/llama/src/test/java/examples/MainExample.java
similarity index 100%
rename from src/test/java/examples/MainExample.java
rename to llama/src/test/java/examples/MainExample.java
diff --git a/src/test/java/examples/OpenAiServerExample.java b/llama/src/test/java/examples/OpenAiServerExample.java
similarity index 100%
rename from src/test/java/examples/OpenAiServerExample.java
rename to llama/src/test/java/examples/OpenAiServerExample.java
diff --git a/src/test/java/net/ladenthin/llama/AudioInputIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/AudioInputIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/AudioInputIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/AudioInputIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/ChatAdvancedTest.java b/llama/src/test/java/net/ladenthin/llama/ChatAdvancedTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ChatAdvancedTest.java
rename to llama/src/test/java/net/ladenthin/llama/ChatAdvancedTest.java
diff --git a/src/test/java/net/ladenthin/llama/ChatScenarioTest.java b/llama/src/test/java/net/ladenthin/llama/ChatScenarioTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ChatScenarioTest.java
rename to llama/src/test/java/net/ladenthin/llama/ChatScenarioTest.java
diff --git a/src/test/java/net/ladenthin/llama/ClaudeGenerated.java b/llama/src/test/java/net/ladenthin/llama/ClaudeGenerated.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ClaudeGenerated.java
rename to llama/src/test/java/net/ladenthin/llama/ClaudeGenerated.java
diff --git a/src/test/java/net/ladenthin/llama/ConfigureParallelInferenceTest.java b/llama/src/test/java/net/ladenthin/llama/ConfigureParallelInferenceTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ConfigureParallelInferenceTest.java
rename to llama/src/test/java/net/ladenthin/llama/ConfigureParallelInferenceTest.java
diff --git a/src/test/java/net/ladenthin/llama/ErrorHandlingTest.java b/llama/src/test/java/net/ladenthin/llama/ErrorHandlingTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ErrorHandlingTest.java
rename to llama/src/test/java/net/ladenthin/llama/ErrorHandlingTest.java
diff --git a/src/test/java/net/ladenthin/llama/LlamaArchitectureTest.java b/llama/src/test/java/net/ladenthin/llama/LlamaArchitectureTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/LlamaArchitectureTest.java
rename to llama/src/test/java/net/ladenthin/llama/LlamaArchitectureTest.java
diff --git a/src/test/java/net/ladenthin/llama/LlamaEmbeddingsTest.java b/llama/src/test/java/net/ladenthin/llama/LlamaEmbeddingsTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/LlamaEmbeddingsTest.java
rename to llama/src/test/java/net/ladenthin/llama/LlamaEmbeddingsTest.java
diff --git a/src/test/java/net/ladenthin/llama/LlamaModelOfflineTest.java b/llama/src/test/java/net/ladenthin/llama/LlamaModelOfflineTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/LlamaModelOfflineTest.java
rename to llama/src/test/java/net/ladenthin/llama/LlamaModelOfflineTest.java
diff --git a/src/test/java/net/ladenthin/llama/LlamaModelTest.java b/llama/src/test/java/net/ladenthin/llama/LlamaModelTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/LlamaModelTest.java
rename to llama/src/test/java/net/ladenthin/llama/LlamaModelTest.java
diff --git a/src/test/java/net/ladenthin/llama/LlamaParameterProperties.java b/llama/src/test/java/net/ladenthin/llama/LlamaParameterProperties.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/LlamaParameterProperties.java
rename to llama/src/test/java/net/ladenthin/llama/LlamaParameterProperties.java
diff --git a/src/test/java/net/ladenthin/llama/MemoryManagementTest.java b/llama/src/test/java/net/ladenthin/llama/MemoryManagementTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/MemoryManagementTest.java
rename to llama/src/test/java/net/ladenthin/llama/MemoryManagementTest.java
diff --git a/src/test/java/net/ladenthin/llama/MultimodalIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/MultimodalIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/MultimodalIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/MultimodalIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/MultimodalMessagesTest.java b/llama/src/test/java/net/ladenthin/llama/MultimodalMessagesTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/MultimodalMessagesTest.java
rename to llama/src/test/java/net/ladenthin/llama/MultimodalMessagesTest.java
diff --git a/src/test/java/net/ladenthin/llama/ReactorIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/ReactorIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ReactorIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/ReactorIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/ReasoningBudgetTest.java b/llama/src/test/java/net/ladenthin/llama/ReasoningBudgetTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ReasoningBudgetTest.java
rename to llama/src/test/java/net/ladenthin/llama/ReasoningBudgetTest.java
diff --git a/src/test/java/net/ladenthin/llama/RerankingModelTest.java b/llama/src/test/java/net/ladenthin/llama/RerankingModelTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/RerankingModelTest.java
rename to llama/src/test/java/net/ladenthin/llama/RerankingModelTest.java
diff --git a/src/test/java/net/ladenthin/llama/ResponseJsonStructureTest.java b/llama/src/test/java/net/ladenthin/llama/ResponseJsonStructureTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ResponseJsonStructureTest.java
rename to llama/src/test/java/net/ladenthin/llama/ResponseJsonStructureTest.java
diff --git a/src/test/java/net/ladenthin/llama/SessionConcurrencyTest.java b/llama/src/test/java/net/ladenthin/llama/SessionConcurrencyTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/SessionConcurrencyTest.java
rename to llama/src/test/java/net/ladenthin/llama/SessionConcurrencyTest.java
diff --git a/src/test/java/net/ladenthin/llama/SessionStateTest.java b/llama/src/test/java/net/ladenthin/llama/SessionStateTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/SessionStateTest.java
rename to llama/src/test/java/net/ladenthin/llama/SessionStateTest.java
diff --git a/src/test/java/net/ladenthin/llama/SessionTest.java b/llama/src/test/java/net/ladenthin/llama/SessionTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/SessionTest.java
rename to llama/src/test/java/net/ladenthin/llama/SessionTest.java
diff --git a/src/test/java/net/ladenthin/llama/TestConstants.java b/llama/src/test/java/net/ladenthin/llama/TestConstants.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/TestConstants.java
rename to llama/src/test/java/net/ladenthin/llama/TestConstants.java
diff --git a/src/test/java/net/ladenthin/llama/ToolCallingAgentTest.java b/llama/src/test/java/net/ladenthin/llama/ToolCallingAgentTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ToolCallingAgentTest.java
rename to llama/src/test/java/net/ladenthin/llama/ToolCallingAgentTest.java
diff --git a/src/test/java/net/ladenthin/llama/ToolCallingIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/ToolCallingIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/ToolCallingIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/ToolCallingIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/TtsIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/TtsIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/TtsIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/TtsIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/AbstractCliArgEnumTest.java b/llama/src/test/java/net/ladenthin/llama/args/AbstractCliArgEnumTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/AbstractCliArgEnumTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/AbstractCliArgEnumTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/CacheTypeTest.java b/llama/src/test/java/net/ladenthin/llama/args/CacheTypeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/CacheTypeTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/CacheTypeTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/ContinuationModeTest.java b/llama/src/test/java/net/ladenthin/llama/args/ContinuationModeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/ContinuationModeTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/ContinuationModeTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/GpuSplitModeTest.java b/llama/src/test/java/net/ladenthin/llama/args/GpuSplitModeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/GpuSplitModeTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/GpuSplitModeTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/LogFormatTest.java b/llama/src/test/java/net/ladenthin/llama/args/LogFormatTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/LogFormatTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/LogFormatTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/MiroStatTest.java b/llama/src/test/java/net/ladenthin/llama/args/MiroStatTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/MiroStatTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/MiroStatTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/ModelFlagTest.java b/llama/src/test/java/net/ladenthin/llama/args/ModelFlagTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/ModelFlagTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/ModelFlagTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/NumaStrategyTest.java b/llama/src/test/java/net/ladenthin/llama/args/NumaStrategyTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/NumaStrategyTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/NumaStrategyTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/PoolingTypeTest.java b/llama/src/test/java/net/ladenthin/llama/args/PoolingTypeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/PoolingTypeTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/PoolingTypeTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/ReasoningFormatTest.java b/llama/src/test/java/net/ladenthin/llama/args/ReasoningFormatTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/ReasoningFormatTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/ReasoningFormatTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/RopeScalingTypeTest.java b/llama/src/test/java/net/ladenthin/llama/args/RopeScalingTypeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/RopeScalingTypeTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/RopeScalingTypeTest.java
diff --git a/src/test/java/net/ladenthin/llama/args/SamplerTest.java b/llama/src/test/java/net/ladenthin/llama/args/SamplerTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/args/SamplerTest.java
rename to llama/src/test/java/net/ladenthin/llama/args/SamplerTest.java
diff --git a/src/test/java/net/ladenthin/llama/benchmark/InferenceParametersBenchmark.java b/llama/src/test/java/net/ladenthin/llama/benchmark/InferenceParametersBenchmark.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/benchmark/InferenceParametersBenchmark.java
rename to llama/src/test/java/net/ladenthin/llama/benchmark/InferenceParametersBenchmark.java
diff --git a/src/test/java/net/ladenthin/llama/callback/CancellationTokenLincheckTest.java b/llama/src/test/java/net/ladenthin/llama/callback/CancellationTokenLincheckTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/callback/CancellationTokenLincheckTest.java
rename to llama/src/test/java/net/ladenthin/llama/callback/CancellationTokenLincheckTest.java
diff --git a/src/test/java/net/ladenthin/llama/callback/CancellationTokenTest.java b/llama/src/test/java/net/ladenthin/llama/callback/CancellationTokenTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/callback/CancellationTokenTest.java
rename to llama/src/test/java/net/ladenthin/llama/callback/CancellationTokenTest.java
diff --git a/src/test/java/net/ladenthin/llama/callback/LoadProgressCallbackTest.java b/llama/src/test/java/net/ladenthin/llama/callback/LoadProgressCallbackTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/callback/LoadProgressCallbackTest.java
rename to llama/src/test/java/net/ladenthin/llama/callback/LoadProgressCallbackTest.java
diff --git a/src/test/java/net/ladenthin/llama/exception/LlamaExceptionTest.java b/llama/src/test/java/net/ladenthin/llama/exception/LlamaExceptionTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/exception/LlamaExceptionTest.java
rename to llama/src/test/java/net/ladenthin/llama/exception/LlamaExceptionTest.java
diff --git a/src/test/java/net/ladenthin/llama/exception/ModelUnavailableExceptionTest.java b/llama/src/test/java/net/ladenthin/llama/exception/ModelUnavailableExceptionTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/exception/ModelUnavailableExceptionTest.java
rename to llama/src/test/java/net/ladenthin/llama/exception/ModelUnavailableExceptionTest.java
diff --git a/src/test/java/net/ladenthin/llama/jcstress/CancellationTokenRace.java b/llama/src/test/java/net/ladenthin/llama/jcstress/CancellationTokenRace.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/jcstress/CancellationTokenRace.java
rename to llama/src/test/java/net/ladenthin/llama/jcstress/CancellationTokenRace.java
diff --git a/src/test/java/net/ladenthin/llama/json/ChatResponseParserTest.java b/llama/src/test/java/net/ladenthin/llama/json/ChatResponseParserTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/json/ChatResponseParserTest.java
rename to llama/src/test/java/net/ladenthin/llama/json/ChatResponseParserTest.java
diff --git a/src/test/java/net/ladenthin/llama/json/ChatStreamChunkParserTest.java b/llama/src/test/java/net/ladenthin/llama/json/ChatStreamChunkParserTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/json/ChatStreamChunkParserTest.java
rename to llama/src/test/java/net/ladenthin/llama/json/ChatStreamChunkParserTest.java
diff --git a/src/test/java/net/ladenthin/llama/json/CompletionResponseParserTest.java b/llama/src/test/java/net/ladenthin/llama/json/CompletionResponseParserTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/json/CompletionResponseParserTest.java
rename to llama/src/test/java/net/ladenthin/llama/json/CompletionResponseParserTest.java
diff --git a/src/test/java/net/ladenthin/llama/json/ParameterJsonSerializerTest.java b/llama/src/test/java/net/ladenthin/llama/json/ParameterJsonSerializerTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/json/ParameterJsonSerializerTest.java
rename to llama/src/test/java/net/ladenthin/llama/json/ParameterJsonSerializerTest.java
diff --git a/src/test/java/net/ladenthin/llama/json/RerankResponseParserTest.java b/llama/src/test/java/net/ladenthin/llama/json/RerankResponseParserTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/json/RerankResponseParserTest.java
rename to llama/src/test/java/net/ladenthin/llama/json/RerankResponseParserTest.java
diff --git a/src/test/java/net/ladenthin/llama/json/TimingsLoggerTest.java b/llama/src/test/java/net/ladenthin/llama/json/TimingsLoggerTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/json/TimingsLoggerTest.java
rename to llama/src/test/java/net/ladenthin/llama/json/TimingsLoggerTest.java
diff --git a/src/test/java/net/ladenthin/llama/loader/LlamaLoaderTest.java b/llama/src/test/java/net/ladenthin/llama/loader/LlamaLoaderTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/loader/LlamaLoaderTest.java
rename to llama/src/test/java/net/ladenthin/llama/loader/LlamaLoaderTest.java
diff --git a/src/test/java/net/ladenthin/llama/loader/LoggingSmokeTest.java b/llama/src/test/java/net/ladenthin/llama/loader/LoggingSmokeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/loader/LoggingSmokeTest.java
rename to llama/src/test/java/net/ladenthin/llama/loader/LoggingSmokeTest.java
diff --git a/src/test/java/net/ladenthin/llama/loader/NativeLibraryLoadSmokeTest.java b/llama/src/test/java/net/ladenthin/llama/loader/NativeLibraryLoadSmokeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/loader/NativeLibraryLoadSmokeTest.java
rename to llama/src/test/java/net/ladenthin/llama/loader/NativeLibraryLoadSmokeTest.java
diff --git a/src/test/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetterTest.java b/llama/src/test/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetterTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetterTest.java
rename to llama/src/test/java/net/ladenthin/llama/loader/NativeLibraryPermissionSetterTest.java
diff --git a/src/test/java/net/ladenthin/llama/loader/OSInfoTest.java b/llama/src/test/java/net/ladenthin/llama/loader/OSInfoTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/loader/OSInfoTest.java
rename to llama/src/test/java/net/ladenthin/llama/loader/OSInfoTest.java
diff --git a/src/test/java/net/ladenthin/llama/parameters/ChatRequestTest.java b/llama/src/test/java/net/ladenthin/llama/parameters/ChatRequestTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/parameters/ChatRequestTest.java
rename to llama/src/test/java/net/ladenthin/llama/parameters/ChatRequestTest.java
diff --git a/src/test/java/net/ladenthin/llama/parameters/InferenceParametersTest.java b/llama/src/test/java/net/ladenthin/llama/parameters/InferenceParametersTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/parameters/InferenceParametersTest.java
rename to llama/src/test/java/net/ladenthin/llama/parameters/InferenceParametersTest.java
diff --git a/src/test/java/net/ladenthin/llama/parameters/JsonEndpointParametersTest.java b/llama/src/test/java/net/ladenthin/llama/parameters/JsonEndpointParametersTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/parameters/JsonEndpointParametersTest.java
rename to llama/src/test/java/net/ladenthin/llama/parameters/JsonEndpointParametersTest.java
diff --git a/src/test/java/net/ladenthin/llama/parameters/JsonParametersTest.java b/llama/src/test/java/net/ladenthin/llama/parameters/JsonParametersTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/parameters/JsonParametersTest.java
rename to llama/src/test/java/net/ladenthin/llama/parameters/JsonParametersTest.java
diff --git a/src/test/java/net/ladenthin/llama/parameters/ModelParametersExtendedTest.java b/llama/src/test/java/net/ladenthin/llama/parameters/ModelParametersExtendedTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/parameters/ModelParametersExtendedTest.java
rename to llama/src/test/java/net/ladenthin/llama/parameters/ModelParametersExtendedTest.java
diff --git a/src/test/java/net/ladenthin/llama/parameters/ModelParametersTest.java b/llama/src/test/java/net/ladenthin/llama/parameters/ModelParametersTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/parameters/ModelParametersTest.java
rename to llama/src/test/java/net/ladenthin/llama/parameters/ModelParametersTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/AnthropicApiSupportTest.java b/llama/src/test/java/net/ladenthin/llama/server/AnthropicApiSupportTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/AnthropicApiSupportTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/AnthropicApiSupportTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/AnthropicStreamTranslatorTest.java b/llama/src/test/java/net/ladenthin/llama/server/AnthropicStreamTranslatorTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/AnthropicStreamTranslatorTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/AnthropicStreamTranslatorTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/NativeServerSmokeTest.java b/llama/src/test/java/net/ladenthin/llama/server/NativeServerSmokeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/NativeServerSmokeTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/NativeServerSmokeTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OaiRerankSupportTest.java b/llama/src/test/java/net/ladenthin/llama/server/OaiRerankSupportTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OaiRerankSupportTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OaiRerankSupportTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OllamaApiSupportTest.java b/llama/src/test/java/net/ladenthin/llama/server/OllamaApiSupportTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OllamaApiSupportTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OllamaApiSupportTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiCompatServerHttpTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiCompatServerHttpTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiCompatServerHttpTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiCompatServerHttpTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiCompatServerIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiCompatServerIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiCompatServerIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiCompatServerIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiRequestMapperTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiRequestMapperTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiRequestMapperTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiRequestMapperTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerCliTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerCliTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerCliTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerCliTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerCompletionIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerCompletionIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerCompletionIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerCompletionIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerConfigTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerConfigTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerConfigTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerConfigTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerEmbeddingsIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerEmbeddingsIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerEmbeddingsIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerEmbeddingsIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerRerankIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerRerankIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerRerankIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerRerankIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerTestSupport.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerTestSupport.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerTestSupport.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerTestSupport.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiServerToolCallingIntegrationTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiServerToolCallingIntegrationTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiServerToolCallingIntegrationTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiServerToolCallingIntegrationTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/OpenAiSseFormatterTest.java b/llama/src/test/java/net/ladenthin/llama/server/OpenAiSseFormatterTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/OpenAiSseFormatterTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/OpenAiSseFormatterTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/ResponsesApiSupportTest.java b/llama/src/test/java/net/ladenthin/llama/server/ResponsesApiSupportTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/ResponsesApiSupportTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/ResponsesApiSupportTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/ResponsesStreamTranslatorTest.java b/llama/src/test/java/net/ladenthin/llama/server/ResponsesStreamTranslatorTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/ResponsesStreamTranslatorTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/ResponsesStreamTranslatorTest.java
diff --git a/src/test/java/net/ladenthin/llama/server/ToolCallDeltaAccumulatorTest.java b/llama/src/test/java/net/ladenthin/llama/server/ToolCallDeltaAccumulatorTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/server/ToolCallDeltaAccumulatorTest.java
rename to llama/src/test/java/net/ladenthin/llama/server/ToolCallDeltaAccumulatorTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ChatChoiceTest.java b/llama/src/test/java/net/ladenthin/llama/value/ChatChoiceTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ChatChoiceTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ChatChoiceTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ChatMessageTest.java b/llama/src/test/java/net/ladenthin/llama/value/ChatMessageTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ChatMessageTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ChatMessageTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ChatResponseTest.java b/llama/src/test/java/net/ladenthin/llama/value/ChatResponseTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ChatResponseTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ChatResponseTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ChatTranscriptTest.java b/llama/src/test/java/net/ladenthin/llama/value/ChatTranscriptTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ChatTranscriptTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ChatTranscriptTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/CompletionResultTest.java b/llama/src/test/java/net/ladenthin/llama/value/CompletionResultTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/CompletionResultTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/CompletionResultTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ContentPartTest.java b/llama/src/test/java/net/ladenthin/llama/value/ContentPartTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ContentPartTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ContentPartTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/LlamaOutputTest.java b/llama/src/test/java/net/ladenthin/llama/value/LlamaOutputTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/LlamaOutputTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/LlamaOutputTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/LogLevelTest.java b/llama/src/test/java/net/ladenthin/llama/value/LogLevelTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/LogLevelTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/LogLevelTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ModelMetaTest.java b/llama/src/test/java/net/ladenthin/llama/value/ModelMetaTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ModelMetaTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ModelMetaTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/PairTest.java b/llama/src/test/java/net/ladenthin/llama/value/PairTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/PairTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/PairTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ServerMetricsTest.java b/llama/src/test/java/net/ladenthin/llama/value/ServerMetricsTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ServerMetricsTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ServerMetricsTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/StopReasonTest.java b/llama/src/test/java/net/ladenthin/llama/value/StopReasonTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/StopReasonTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/StopReasonTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/TimingsTest.java b/llama/src/test/java/net/ladenthin/llama/value/TimingsTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/TimingsTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/TimingsTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/TokenLogprobTest.java b/llama/src/test/java/net/ladenthin/llama/value/TokenLogprobTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/TokenLogprobTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/TokenLogprobTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ToolCallTest.java b/llama/src/test/java/net/ladenthin/llama/value/ToolCallTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ToolCallTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ToolCallTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/ToolDefinitionTest.java b/llama/src/test/java/net/ladenthin/llama/value/ToolDefinitionTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/ToolDefinitionTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/ToolDefinitionTest.java
diff --git a/src/test/java/net/ladenthin/llama/value/UsageTest.java b/llama/src/test/java/net/ladenthin/llama/value/UsageTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/value/UsageTest.java
rename to llama/src/test/java/net/ladenthin/llama/value/UsageTest.java
diff --git a/src/test/java/net/ladenthin/llama/vmlens/SessionStateInterleavingTest.java b/llama/src/test/java/net/ladenthin/llama/vmlens/SessionStateInterleavingTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/vmlens/SessionStateInterleavingTest.java
rename to llama/src/test/java/net/ladenthin/llama/vmlens/SessionStateInterleavingTest.java
diff --git a/src/test/java/net/ladenthin/llama/vmlens/VmlensInterleavingSmokeTest.java b/llama/src/test/java/net/ladenthin/llama/vmlens/VmlensInterleavingSmokeTest.java
similarity index 100%
rename from src/test/java/net/ladenthin/llama/vmlens/VmlensInterleavingSmokeTest.java
rename to llama/src/test/java/net/ladenthin/llama/vmlens/VmlensInterleavingSmokeTest.java
diff --git a/src/test/resources/images/README.md b/llama/src/test/resources/images/README.md
similarity index 100%
rename from src/test/resources/images/README.md
rename to llama/src/test/resources/images/README.md
diff --git a/src/test/resources/images/test-image.jpg b/llama/src/test/resources/images/test-image.jpg
similarity index 100%
rename from src/test/resources/images/test-image.jpg
rename to llama/src/test/resources/images/test-image.jpg
diff --git a/pom.xml b/pom.xml
index d8c1492c..c73ba70d 100644
--- a/pom.xml
+++ b/pom.xml
@@ -10,14 +10,21 @@ SPDX-License-Identifier: MIT
 	xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
 	<modelVersion>4.0.0</modelVersion>
 
+	<!--
+		Aggregator / parent POM. Carries no code — it only groups the modules and holds the
+		single shared <version>, which every module inherits (so the core net.ladenthin:llama
+		and net.ladenthin:llama-langchain4j always ship in lockstep by construction). The core
+		artifact keeps its coordinates net.ladenthin:llama (the `llama` module); consumers are
+		unaffected by the directory move.
+	-->
 	<groupId>net.ladenthin</groupId>
-	<artifactId>llama</artifactId>
+	<artifactId>llama-parent</artifactId>
 	<version>5.0.4-SNAPSHOT</version>
-	<packaging>jar</packaging>
+	<packaging>pom</packaging>
 
 	<name>${project.groupId}:${project.artifactId}</name>
-	<description>Java Bindings for llama.cpp - A Port of Facebook's LLaMA model
-		in C/C++.</description>
+	<description>Aggregator POM for java-llama.cpp: the native JNI binding
+		(net.ladenthin:llama) and its sibling integrations.</description>
 	<url>https://github.com/bernardladenthin/java-llama.cpp</url>
 
 	<licenses>
@@ -48,671 +55,20 @@ SPDX-License-Identifier: MIT
 		</snapshotRepository>
 	</distributionManagement>
 
-	<properties>
-		<sonar.organization>bernardladenthin</sonar.organization>
-		<jna.version>5.19.0</jna.version>
-		<jspecify.version>1.0.0</jspecify.version>
-		<lombok.version>1.18.46</lombok.version>
-		<errorprone.version>2.50.0</errorprone.version>
-		<nullaway.version>0.13.7</nullaway.version>
-		<checker.version>4.2.0</checker.version>
-		<jackson.version>2.22.0</jackson.version>
-		<reactor.version>3.8.6</reactor.version>
-		<slf4j.version>2.0.18</slf4j.version>
-		<logback.version>1.5.37</logback.version>
-		<animal-sniffer.version>1.27</animal-sniffer.version>
-		<junit.version>6.1.1</junit.version>
-		<hamcrest.version>3.0</hamcrest.version>
-		<jmh.version>1.37</jmh.version>
-		<jcstress.version>0.16</jcstress.version>
-		<lincheck.version>3.6</lincheck.version>
-		<logcaptor.version>2.12.6</logcaptor.version>
-		<vmlens.version>1.2.28</vmlens.version>
-		<!-- DO NOT UPGRADE jqwik past 1.9.3. jqwik 1.10.0 added a deliberate
-		     anti-AI prompt-injection string to test stdout; the 1.10.1 user
-		     guide states the library "is not meant to be used by any 'AI'
-		     coding agents at all." 1.9.3 is the last pre-disclosure release
-		     and is the pinned version for this repo. Any CI / Dependabot /
-		     contributor PR that bumps this MUST be rejected. See CLAUDE.md
-		     section "jqwik prompt-injection in test output" for full context. -->
-		<jqwik.version>1.9.3</jqwik.version>
-		<archunit.version>1.4.2</archunit.version>
-		<spotbugs.version>4.10.2.0</spotbugs.version>
-		<fb-contrib.version>7.7.4</fb-contrib.version>
-		<findsecbugs.version>1.14.0</findsecbugs.version>
-		<spotless.version>3.7.0</spotless.version>
-		<palantir-java-format.version>2.94.0</palantir-java-format.version>
-		<project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-		<project.build.outputTimestamp>${git.commit.time}</project.build.outputTimestamp>
-	</properties>
+	<modules>
+		<module>llama</module>
+		<module>llama-langchain4j</module>
+	</modules>
 
-	<!--
-	    Pin transitively-conflicting versions so maven-enforcer's
-	    dependencyConvergence rule passes. The direct slf4j-api version above
-	    must always win over the older 2.0.17 that logback-classic 1.5.32
-	    brings in transitively.
-	-->
-	<dependencyManagement>
-		<dependencies>
-			<dependency>
-				<groupId>org.slf4j</groupId>
-				<artifactId>slf4j-api</artifactId>
-				<version>${slf4j.version}</version>
-			</dependency>
-		</dependencies>
-	</dependencyManagement>
-
-	<dependencies>
-		<dependency>
-			<groupId>org.projectlombok</groupId>
-			<artifactId>lombok</artifactId>
-			<version>${lombok.version}</version>
-			<scope>provided</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.junit.jupiter</groupId>
-			<artifactId>junit-jupiter</artifactId>
-			<version>${junit.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.hamcrest</groupId>
-			<artifactId>hamcrest</artifactId>
-			<version>${hamcrest.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>net.jqwik</groupId>
-			<artifactId>jqwik</artifactId>
-			<version>${jqwik.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>com.tngtech.archunit</groupId>
-			<artifactId>archunit-junit5</artifactId>
-			<version>${archunit.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.jspecify</groupId>
-			<artifactId>jspecify</artifactId>
-			<version>${jspecify.version}</version>
-		</dependency>
-		<dependency>
-			<groupId>org.checkerframework</groupId>
-			<artifactId>checker-qual</artifactId>
-			<version>${checker.version}</version>
-		</dependency>
-		<dependency>
-			<groupId>com.fasterxml.jackson.core</groupId>
-			<artifactId>jackson-databind</artifactId>
-			<version>${jackson.version}</version>
-		</dependency>
-		<!-- Required by OSInfo (vendored from xerial/sqlite-jdbc) for log emission. -->
-		<dependency>
-			<groupId>org.slf4j</groupId>
-			<artifactId>slf4j-api</artifactId>
-			<version>${slf4j.version}</version>
-		</dependency>
-		<!-- Default SLF4J binding shipped with this library. Runtime scope: not
-		     required on the compile classpath, only loaded at JVM startup. -->
-		<dependency>
-			<groupId>ch.qos.logback</groupId>
-			<artifactId>logback-classic</artifactId>
-			<version>${logback.version}</version>
-			<scope>runtime</scope>
-		</dependency>
-		<!-- @IgnoreJRERequirement marker used by OSInfo (vendored from xerial/sqlite-jdbc)
-		     on a few methods that touch APIs unavailable on Android. Retention=CLASS, so
-		     the dep is compile-only (provided scope). -->
-		<dependency>
-			<groupId>org.codehaus.mojo</groupId>
-			<artifactId>animal-sniffer-annotations</artifactId>
-			<version>${animal-sniffer.version}</version>
-			<scope>provided</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.openjdk.jmh</groupId>
-			<artifactId>jmh-core</artifactId>
-			<version>${jmh.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.openjdk.jmh</groupId>
-			<artifactId>jmh-generator-annprocess</artifactId>
-			<version>${jmh.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.openjdk.jcstress</groupId>
-			<artifactId>jcstress-core</artifactId>
-			<version>${jcstress.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>org.jetbrains.lincheck</groupId>
-			<artifactId>lincheck</artifactId>
-			<version>${lincheck.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<!-- Test-only SLF4J/Logback log-capture for the LoggingSmokeTest. -->
-		<dependency>
-			<groupId>io.github.hakky54</groupId>
-			<artifactId>logcaptor</artifactId>
-			<version>${logcaptor.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<!-- Test-only Project Reactor used by ReactorIntegrationTest to prove the
-		     "wrap LlamaIterable with Flux.fromIterable + subscribeOn" pattern works
-		     end-to-end (backpressure, AutoCloseable cancel propagation). NOT a runtime
-		     dependency — consumers bring whichever reactive lib they already use
-		     (Reactor / RxJava / Kotlin Flow / Akka — see README "Reactive integration"). -->
-		<dependency>
-			<groupId>io.projectreactor</groupId>
-			<artifactId>reactor-core</artifactId>
-			<version>${reactor.version}</version>
-			<scope>test</scope>
-		</dependency>
-		<dependency>
-			<groupId>io.projectreactor</groupId>
-			<artifactId>reactor-test</artifactId>
-			<version>${reactor.version}</version>
-			<scope>test</scope>
-		</dependency>
+	<profiles>
 		<!--
-			vmlens interleaving-analysis API. Test-scoped and transitive-dependency-free
-			(its own deps are all test-scope, hence not propagated), so it is safe on the
-			default test classpath under dependencyConvergence. Needed so the
-			VmlensInterleavingSmoke* test compiles in every build; it is only *executed*
-			under the `vmlens` profile (see that profile + the vmlens CI job). The default
-			surefire run excludes it (managed surefire <excludes> below) so the vmlens
-			agent's JDK 9 API class is never loaded outside the agent-driven run.
+			Reactor-wide release profile. Inherited by every module (llama, llama-langchain4j)
+			AND applied to this parent pom itself, so a single `mvn -P release deploy` GPG-signs
+			all artifacts (including the parent .pom, which Central requires) and routes the whole
+			reactor through the Central Publishing plugin as one deployment. Activated only by the
+			publish-snapshot / publish-release CI jobs. Plugin versions are pinned here explicitly
+			(the parent has no pluginManagement to inherit them from).
 		-->
-		<dependency>
-			<groupId>com.vmlens</groupId>
-			<artifactId>api</artifactId>
-			<version>${vmlens.version}</version>
-			<scope>test</scope>
-		</dependency>
-	</dependencies>
-
-	<build>
-		<pluginManagement>
-			<plugins>
-				<plugin>
-					<groupId>com.diffplug.spotless</groupId>
-					<artifactId>spotless-maven-plugin</artifactId>
-					<version>${spotless.version}</version>
-				</plugin>
-				<plugin>
-					<groupId>com.github.spotbugs</groupId>
-					<artifactId>spotbugs-maven-plugin</artifactId>
-					<version>${spotbugs.version}</version>
-				</plugin>
-				<plugin>
-					<groupId>com.vmlens</groupId>
-					<artifactId>vmlens-maven-plugin</artifactId>
-					<version>${vmlens.version}</version>
-				</plugin>
-				<plugin>
-					<groupId>io.github.git-commit-id</groupId>
-					<artifactId>git-commit-id-maven-plugin</artifactId>
-					<version>10.0.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-assembly-plugin</artifactId>
-					<version>3.8.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-compiler-plugin</artifactId>
-					<version>3.15.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-gpg-plugin</artifactId>
-					<version>3.2.8</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-jar-plugin</artifactId>
-					<version>3.5.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-javadoc-plugin</artifactId>
-					<version>3.12.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-resources-plugin</artifactId>
-					<version>3.5.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-source-plugin</artifactId>
-					<version>3.4.0</version>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-surefire-plugin</artifactId>
-					<version>3.5.6</version>
-					<configuration>
-						<!--
-							Tests in the `vmlens` package are meaningful only when run
-							under the vmlens agent (the `vmlens` profile / CI job). Without the
-							agent its AllInterleavings loop body never executes (a vacuous pass
-							that also prints an "agent not configured" warning), so exclude it
-							from the ordinary surefire run. The vmlens job re-includes them with
-							`-Dtest=...`, which overrides this exclude.
-						-->
-						<excludes>
-							<exclude>**/vmlens/*.java</exclude>
-						</excludes>
-					</configuration>
-				</plugin>
-				<plugin>
-					<groupId>org.apache.maven.plugins</groupId>
-					<artifactId>maven-enforcer-plugin</artifactId>
-					<version>3.6.3</version>
-				</plugin>
-				<plugin>
-					<groupId>org.codehaus.mojo</groupId>
-					<artifactId>exec-maven-plugin</artifactId>
-					<version>3.6.3</version>
-				</plugin>
-				<plugin>
-					<groupId>org.jacoco</groupId>
-					<artifactId>jacoco-maven-plugin</artifactId>
-					<version>0.8.15</version>
-				</plugin>
-				<plugin>
-					<groupId>org.pitest</groupId>
-					<artifactId>pitest-maven</artifactId>
-					<version>1.25.5</version>
-				</plugin>
-				<plugin>
-					<groupId>org.sonatype.central</groupId>
-					<artifactId>central-publishing-maven-plugin</artifactId>
-					<version>0.11.0</version>
-				</plugin>
-			</plugins>
-		</pluginManagement>
-		<plugins>
-			<plugin>
-				<groupId>org.apache.maven.plugins</groupId>
-				<artifactId>maven-enforcer-plugin</artifactId>
-				<executions>
-					<execution>
-						<id>enforce</id>
-						<goals>
-							<goal>enforce</goal>
-						</goals>
-						<configuration>
-							<rules>
-								<requireMavenVersion>
-									<version>[3.6.3,)</version>
-								</requireMavenVersion>
-								<requireJavaVersion>
-									<version>[1.8,)</version>
-								</requireJavaVersion>
-								<dependencyConvergence/>
-								<bannedDependencies>
-									<excludes>
-										<!-- Use SLF4J directly; never commons-logging. -->
-										<exclude>commons-logging:commons-logging</exclude>
-										<!-- Legacy / EoL logging frameworks. -->
-										<exclude>log4j:log4j</exclude>
-										<!-- Old hamcrest split artifacts; use hamcrest:hamcrest 2.x+. -->
-										<exclude>org.hamcrest:hamcrest-core</exclude>
-										<exclude>org.hamcrest:hamcrest-library</exclude>
-										<exclude>org.hamcrest:hamcrest-all</exclude>
-										<!-- Legacy JUnit 3 / 4; we use Jupiter (junit:junit-jupiter). -->
-										<exclude>junit:junit</exclude>
-										<exclude>junit:junit-dep</exclude>
-									</excludes>
-								</bannedDependencies>
-							</rules>
-						</configuration>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<groupId>io.github.git-commit-id</groupId>
-				<artifactId>git-commit-id-maven-plugin</artifactId>
-				<executions>
-					<execution>
-						<id>get-git-properties</id>
-						<goals>
-							<goal>revision</goal>
-						</goals>
-						<phase>initialize</phase>
-					</execution>
-				</executions>
-				<configuration>
-					<dateFormat>yyyy-MM-dd'T'HH:mm:ss'Z'</dateFormat>
-					<dateFormatTimeZone>UTC</dateFormatTimeZone>
-					<generateGitPropertiesFile>false</generateGitPropertiesFile>
-					<failOnNoGitDirectory>false</failOnNoGitDirectory>
-				</configuration>
-			</plugin>
-			<plugin>
-				<groupId>org.apache.maven.plugins</groupId>
-				<artifactId>maven-compiler-plugin</artifactId>
-				<configuration>
-					<release>8</release>
-					<testRelease>21</testRelease>
-					<showWarnings>true</showWarnings>
-					<parameters>true</parameters>
-					<compilerArgs>
-						<!--
-						  -Xlint:all surfaces every category of javac warning. Excluded:
-						    -serial    : serialVersionUID warnings on non-Serializable classes
-						    -options   : bootclasspath / target-mismatch chatter under release=8
-						    -classfile : noise when javac reads our own .class output mid-build
-						    -processing: "No processor claimed any of these annotations" — the
-						                 animal-sniffer @IgnoreJRERequirement is consumed by the
-						                 maven-plugin (not a JSR-269 processor), so javac flags
-						                 it every build
-						  -Werror is on: the ElementType.MODULE blocker is gone now that
-						  @NullMarked lives only in module-info.java (compiled at release 9 in
-						  a separate execution); see module-info.java's javadoc.
-						-->
-						<arg>-Xlint:all,-serial,-options,-classfile,-processing</arg>
-						<arg>-Werror</arg>
-						<!--
-						  Checker Framework Nullness Checker runs as a second nullness pass
-						  alongside NullAway. CF is generics-aware and stricter than NullAway,
-						  so it acts as a second-opinion verifier on the same JSpecify
-						  annotations.
-						-->
-						<!-- Explicit processor list: Lombok (two SPI classes) + Checker Framework's
-							 Nullness Checker as a 2nd nullness pass alongside NullAway. The explicit
-							 -processor argument overrides SPI discovery, so every processor we want
-							 to run must be listed here. -->
-						<arg>-processor</arg>
-						<arg>lombok.launch.AnnotationProcessorHider$AnnotationProcessor,lombok.launch.AnnotationProcessorHider$ClaimingProcessor,org.checkerframework.checker.nullness.NullnessChecker</arg>
-						<arg>-XDaddTypeAnnotationsToSymbol=true</arg>
-						<arg>-XDcompilePolicy=simple</arg>
-						<arg>--should-stop=ifError=FLOW</arg>
-						<arg>-Xplugin:ErrorProne -Xep:NullAway:ERROR -XepOpt:NullAway:OnlyNullMarked=true -XepOpt:NullAway:JSpecifyMode=true -XepOpt:NullAway:CheckOptionalEmptiness=true -XepOpt:NullAway:AcknowledgeRestrictiveAnnotations=true -XepOpt:NullAway:AcknowledgeAndroidRecent=true -XepOpt:NullAway:AssertsEnabled=true -Xep:BoxedPrimitiveEquality:ERROR -Xep:EqualsHashCode:ERROR -Xep:EqualsIncompatibleType:ERROR -Xep:IdentityBinaryExpression:ERROR -Xep:SelfAssignment:ERROR -Xep:SelfComparison:ERROR -Xep:SelfEquals:ERROR -Xep:DeadException:ERROR -Xep:FormatString:ERROR -Xep:InvalidPatternSyntax:ERROR -Xep:OptionalEquality:ERROR -Xep:ImpossibleNullComparison:ERROR</arg>
-					</compilerArgs>
-					<annotationProcessorPaths>
-						<path>
-							<groupId>org.projectlombok</groupId>
-							<artifactId>lombok</artifactId>
-							<version>${lombok.version}</version>
-						</path>
-						<path>
-							<groupId>com.google.errorprone</groupId>
-							<artifactId>error_prone_core</artifactId>
-							<version>${errorprone.version}</version>
-						</path>
-						<path>
-							<groupId>com.uber.nullaway</groupId>
-							<artifactId>nullaway</artifactId>
-							<version>${nullaway.version}</version>
-						</path>
-						<path>
-							<groupId>org.checkerframework</groupId>
-							<artifactId>checker</artifactId>
-							<version>${checker.version}</version>
-						</path>
-					</annotationProcessorPaths>
-				</configuration>
-				<executions>
-					<execution>
-						<id>default-compile</id>
-						<configuration>
-							<!-- The default compile pass targets Java 8 bytecode for the library
-							     sources. module-info.java cannot be compiled at release 8 (it is a
-							     Java 9+ construct) and is handled by the separate execution below. -->
-							<excludes>
-								<exclude>module-info.java</exclude>
-							</excludes>
-						</configuration>
-					</execution>
-					<execution>
-						<id>module-info-compile</id>
-						<phase>compile</phase>
-						<goals>
-							<goal>compile</goal>
-						</goals>
-						<configuration>
-							<!-- module-info.java is the only Java 9+ source in this project; compile
-							     it at release 9 (the minimum version that understands modules) without
-							     the Error Prone / NullAway / Checker Framework processors that target
-							     ordinary source files only. Java 8 runtimes silently ignore the
-							     resulting module-info.class at the JAR root, so existing classpath
-							     consumers continue to work unchanged. LlamaLoader loads native libs
-							     via Class.getResourceAsStream on its own class object, which resolves
-							     to this module and therefore does not require any opens directives. -->
-							<release>9</release>
-							<includes>
-								<include>module-info.java</include>
-							</includes>
-							<compilerArgs combine.self="override"/>
-							<annotationProcessorPaths combine.self="override"/>
-						</configuration>
-					</execution>
-					<execution>
-						<id>default-testCompile</id>
-						<configuration>
-							<!-- Drop -parameters for test compile so jcstress reflection doesn't
-							     pick up parameter names from MethodParameters. -->
-							<parameters>false</parameters>
-							<compilerArgs combine.self="override">
-								<arg>-XDaddTypeAnnotationsToSymbol=true</arg>
-								<arg>-XDcompilePolicy=simple</arg>
-								<arg>--should-stop=ifError=FLOW</arg>
-								<arg>-Xplugin:ErrorProne -Xep:NullAway:OFF -Xep:GuardedBy:OFF</arg>
-							</compilerArgs>
-							<annotationProcessorPaths combine.children="append">
-								<path>
-									<groupId>org.openjdk.jcstress</groupId>
-									<artifactId>jcstress-core</artifactId>
-									<version>${jcstress.version}</version>
-								</path>
-								<path>
-									<groupId>org.openjdk.jmh</groupId>
-									<artifactId>jmh-generator-annprocess</artifactId>
-									<version>${jmh.version}</version>
-								</path>
-							</annotationProcessorPaths>
-						</configuration>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<artifactId>maven-resources-plugin</artifactId>
-			</plugin>
-			<plugin>
-				<groupId>org.apache.maven.plugins</groupId>
-				<artifactId>maven-source-plugin</artifactId>
-				<executions>
-					<execution>
-						<id>attach-sources</id>
-						<goals>
-							<goal>jar-no-fork</goal>
-						</goals>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<groupId>org.apache.maven.plugins</groupId>
-				<artifactId>maven-javadoc-plugin</artifactId>
-				<configuration>
-					<source>1.8</source>
-					<failOnError>true</failOnError>
-					<failOnWarnings>true</failOnWarnings>
-					<doclint>all</doclint>
-				</configuration>
-				<executions>
-					<execution>
-						<id>attach-javadocs</id>
-						<goals>
-							<goal>jar</goal>
-						</goals>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<groupId>org.jacoco</groupId>
-				<artifactId>jacoco-maven-plugin</artifactId>
-				<executions>
-					<execution>
-						<id>prepare-agent</id>
-						<goals>
-							<goal>prepare-agent</goal>
-						</goals>
-					</execution>
-					<execution>
-						<id>report</id>
-						<phase>test</phase>
-						<goals>
-							<goal>report</goal>
-						</goals>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<!--
-				  Surefire is configured here only to control native-crash diagnostics:
-				  - argLine inherits jacoco's @{argLine} (so the jacoco runtime agent stays
-				    attached) and appends -XX:ErrorFile so the JVM crash log (hs_err_pid<pid>.log)
-				    always lands in the project basedir, which is the same directory the CI
-				    upload-artifact globs (${{ github.workspace }}/hs_err_pid*.log).
-				  - HeapDumpOnOutOfMemoryError + HeapDumpPath catch OOMs that would otherwise
-				    just print a stack and exit.
-				  - The version is pinned to the same release Maven would otherwise auto-select
-				    (visible in CI error logs as 3.5.4), so we don't drift independently of
-				    Maven's plugin-defaults bump.
-				-->
-				<groupId>org.apache.maven.plugins</groupId>
-				<artifactId>maven-surefire-plugin</artifactId>
-				<configuration>
-					<argLine>@{argLine} -Xmx2g -XX:ErrorFile=hs_err_pid%p.log -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=.</argLine>
-					<!--
-					  Capture each test class's stdout/stderr into
-					  target/surefire-reports/<class>-output.txt. When a native crash
-					  takes down the forked JVM, this file is the ONLY record of
-					  which @Test method was last running (without it the surefire
-					  *.txt summary just says "VM terminated", which we already saw
-					  in CI without telling us which test was on fire).
-					-->
-					<redirectTestOutputToFile>true</redirectTestOutputToFile>
-				</configuration>
-			</plugin>
-			<plugin>
-				<groupId>com.diffplug.spotless</groupId>
-				<artifactId>spotless-maven-plugin</artifactId>
-				<configuration>
-					<java>
-						<includes>
-							<include>src/main/java/**/*.java</include>
-							<include>src/test/java/**/*.java</include>
-						</includes>
-						<palantirJavaFormat>
-							<version>${palantir-java-format.version}</version>
-						</palantirJavaFormat>
-						<removeUnusedImports/>
-						<trimTrailingWhitespace/>
-						<endWithNewline/>
-					</java>
-				</configuration>
-				<executions>
-					<execution>
-						<id>spotless-check</id>
-						<phase>verify</phase>
-						<goals>
-							<goal>check</goal>
-						</goals>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<groupId>com.github.spotbugs</groupId>
-				<artifactId>spotbugs-maven-plugin</artifactId>
-				<configuration>
-					<effort>Max</effort>
-					<threshold>Low</threshold>
-					<failOnError>true</failOnError>
-					<includeTests>false</includeTests>
-					<excludeFilterFile>spotbugs-exclude.xml</excludeFilterFile>
-					<plugins>
-						<plugin>
-							<groupId>com.mebigfatguy.fb-contrib</groupId>
-							<artifactId>fb-contrib</artifactId>
-							<version>${fb-contrib.version}</version>
-						</plugin>
-						<plugin>
-							<groupId>com.h3xstream.findsecbugs</groupId>
-							<artifactId>findsecbugs-plugin</artifactId>
-							<version>${findsecbugs.version}</version>
-						</plugin>
-					</plugins>
-				</configuration>
-				<executions>
-					<execution>
-						<id>spotbugs-check</id>
-						<phase>verify</phase>
-						<goals>
-							<goal>check</goal>
-						</goals>
-					</execution>
-				</executions>
-			</plugin>
-			<plugin>
-				<groupId>org.codehaus.mojo</groupId>
-				<artifactId>exec-maven-plugin</artifactId>
-				<configuration>
-					<mainClass>org.openjdk.jmh.Main</mainClass>
-					<classpathScope>test</classpathScope>
-				</configuration>
-			</plugin>
-			<plugin>
-				<!--
-				  Generic PIT policy (version, 100% gate, invocation, expansion rule):
-				  see ../workspace/policies/pit-mutation-testing.md.
-				  Repo-specific: the gated value/exception/args/json classes are pure-Java
-				  (no native libjllama or model file needed); targetClasses/targetTests are
-				  kept in lock-step. NOTE: value.ContentPart.audioFile(Path) mutants are only
-				  killed with the audio fixture present — see TODO.md and policy §4.
-				-->
-				<groupId>org.pitest</groupId>
-				<artifactId>pitest-maven</artifactId>
-				<dependencies>
-					<dependency>
-						<groupId>org.pitest</groupId>
-						<artifactId>pitest-junit5-plugin</artifactId>
-						<version>1.2.3</version>
-					</dependency>
-				</dependencies>
-				<configuration>
-					<targetClasses>
-						<param>net.ladenthin.llama.value.*</param>
-						<param>net.ladenthin.llama.exception.*</param>
-						<param>net.ladenthin.llama.args.*</param>
-						<param>net.ladenthin.llama.json.TimingsLogger</param>
-						<param>net.ladenthin.llama.json.RerankResponseParser</param>
-						<param>net.ladenthin.llama.json.ChatResponseParser</param>
-						<param>net.ladenthin.llama.json.CompletionResponseParser</param>
-					</targetClasses>
-					<targetTests>
-						<param>net.ladenthin.llama.value.*</param>
-						<param>net.ladenthin.llama.exception.*</param>
-						<param>net.ladenthin.llama.args.*</param>
-						<param>net.ladenthin.llama.json.*</param>
-					</targetTests>
-					<mutationThreshold>100</mutationThreshold>
-					<timeoutConstant>30000</timeoutConstant>
-				</configuration>
-			</plugin>
-		</plugins>
-	</build>
-
-	<profiles>
 		<profile>
 			<id>release</id>
 			<build>
@@ -720,6 +76,7 @@ SPDX-License-Identifier: MIT
 					<plugin>
 						<groupId>org.apache.maven.plugins</groupId>
 						<artifactId>maven-gpg-plugin</artifactId>
+						<version>3.2.8</version>
 						<executions>
 							<execution>
 								<id>sign-artifacts</id>
@@ -739,6 +96,7 @@ SPDX-License-Identifier: MIT
 					<plugin>
 						<groupId>org.sonatype.central</groupId>
 						<artifactId>central-publishing-maven-plugin</artifactId>
+						<version>0.11.0</version>
 						<extensions>true</extensions>
 						<configuration>
 							<publishingServerId>central</publishingServerId>
@@ -749,601 +107,5 @@ SPDX-License-Identifier: MIT
 				</plugins>
 			</build>
 		</profile>
-
-		<profile>
-			<id>cuda</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-compiler-plugin</artifactId>
-						<executions>
-						<!-- We have to perform a separate build pass for cuda
-						classifier -->
-						<execution>
-							<id>gpu</id>
-							<phase>compile</phase>
-							<goals>
-								<goal>compile</goal>
-							</goals>
-							<configuration>
-								<!-- Same rationale as default-compile: this pass targets release 8;
-								     module-info.java is Java 9+ and is handled by module-info-compile. -->
-								<excludes>
-									<exclude>module-info.java</exclude>
-								</excludes>
-								<compilerArgs>
-									<arg>-h</arg>
-									<arg>src/main/cpp</arg>
-								</compilerArgs>
-								<outputDirectory>
-									${project.build.outputDirectory}_cuda</outputDirectory>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<artifactId>maven-resources-plugin</artifactId>
-						<executions>
-						<!-- Copy custom cuda libs to the output directory -->
-						<execution>
-							<id>copy-resources</id>
-							<phase>process-classes</phase>
-							<goals>
-								<goal>copy-resources</goal>
-							</goals>
-							<configuration>
-								<outputDirectory>
-									${project.build.outputDirectory}_cuda</outputDirectory>
-								<resources>
-									<resource>
-										<directory>
-											${basedir}/src/main/resources_linux_cuda/</directory>
-										<includes>
-											<include>**/*.*</include>
-										</includes>
-									</resource>
-								</resources>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-jar-plugin</artifactId>
-						<executions>
-							<!-- Pick class files AND libs from custom output
-							directory -->
-							<execution>
-								<id>cuda</id>
-								<phase>package</phase>
-								<goals>
-									<goal>jar</goal>
-								</goals>
-								<configuration>
-									<classifier>cuda13-linux-x86-64</classifier>
-									<classesDirectory>
-										${project.build.outputDirectory}_cuda</classesDirectory>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-
-		<profile>
-			<id>opencl-android</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-compiler-plugin</artifactId>
-						<executions>
-						<!-- Separate build pass for opencl-android classifier -->
-						<execution>
-							<id>opencl-android</id>
-							<phase>compile</phase>
-							<goals>
-								<goal>compile</goal>
-							</goals>
-							<configuration>
-								<!-- Same rationale as default-compile: this pass targets release 8;
-								     module-info.java is Java 9+ and is handled by module-info-compile. -->
-								<excludes>
-									<exclude>module-info.java</exclude>
-								</excludes>
-								<compilerArgs>
-									<arg>-h</arg>
-									<arg>src/main/cpp</arg>
-								</compilerArgs>
-								<outputDirectory>
-									${project.build.outputDirectory}_opencl_android</outputDirectory>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<artifactId>maven-resources-plugin</artifactId>
-						<executions>
-						<!-- Copy custom opencl-android libs to the output directory -->
-						<execution>
-							<id>copy-resources-opencl-android</id>
-							<phase>process-classes</phase>
-							<goals>
-								<goal>copy-resources</goal>
-							</goals>
-							<configuration>
-								<outputDirectory>
-									${project.build.outputDirectory}_opencl_android</outputDirectory>
-								<resources>
-									<resource>
-										<directory>
-											${basedir}/src/main/resources_android_opencl/</directory>
-										<includes>
-											<include>**/*.*</include>
-										</includes>
-									</resource>
-								</resources>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-jar-plugin</artifactId>
-						<executions>
-							<!-- Pick class files AND libs from custom output
-							directory -->
-							<execution>
-								<id>opencl-android</id>
-								<phase>package</phase>
-								<goals>
-									<goal>jar</goal>
-								</goals>
-								<configuration>
-									<classifier>opencl-android-aarch64</classifier>
-									<classesDirectory>
-										${project.build.outputDirectory}_opencl_android</classesDirectory>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-		<profile>
-			<!-- MSVC (Visual Studio generator) Windows CPU natives, shipped as the
-			     `msvc-windows` classifier JAR. After the default flip the DEFAULT JAR
-			     ships the Ninja Multi-Config natives (sccache-cached, merged into
-			     src/main/resources/ by CI); this profile packages the alternative
-			     MSVC-generator build for anyone who wants it. Both use the same MSVC
-			     toolchain (static /MT CRT) so they are functionally equivalent. Mirrors
-			     the cuda / opencl-android classifier pattern: a separate compile pass +
-			     resource copy + classified jar. The MSVC DLLs are staged into
-			     src/main/resources_windows_msvc/ by CI before this profile runs. -->
-			<id>windows-msvc</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-compiler-plugin</artifactId>
-						<executions>
-						<!-- Separate build pass for the windows-msvc classifier -->
-						<execution>
-							<id>windows-msvc</id>
-							<phase>compile</phase>
-							<goals>
-								<goal>compile</goal>
-							</goals>
-							<configuration>
-								<!-- Same rationale as default-compile: this pass targets release 8;
-								     module-info.java is Java 9+ and is handled by module-info-compile. -->
-								<excludes>
-									<exclude>module-info.java</exclude>
-								</excludes>
-								<compilerArgs>
-									<arg>-h</arg>
-									<arg>src/main/cpp</arg>
-								</compilerArgs>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_msvc</outputDirectory>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<artifactId>maven-resources-plugin</artifactId>
-						<executions>
-						<!-- Copy the Ninja-built Windows libs to the output directory -->
-						<execution>
-							<id>copy-resources-windows-msvc</id>
-							<phase>process-classes</phase>
-							<goals>
-								<goal>copy-resources</goal>
-							</goals>
-							<configuration>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_msvc</outputDirectory>
-								<resources>
-									<resource>
-										<directory>
-											${basedir}/src/main/resources_windows_msvc/</directory>
-										<includes>
-											<include>**/*.*</include>
-										</includes>
-									</resource>
-								</resources>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-jar-plugin</artifactId>
-						<executions>
-							<!-- Pick class files AND libs from custom output
-							directory -->
-							<execution>
-								<id>windows-msvc</id>
-								<phase>package</phase>
-								<goals>
-									<goal>jar</goal>
-								</goals>
-								<configuration>
-									<classifier>msvc-windows</classifier>
-									<classesDirectory>
-										${project.build.outputDirectory}_windows_msvc</classesDirectory>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-
-		<profile>
-			<!-- Windows CUDA GPU natives (x86_64 only), shipped as the
-			     `cuda13-windows-x86-64` classifier JAR. Built with -DGGML_CUDA=ON and
-			     routed by CMakeLists.txt to src/main/resources_windows_cuda/. The CUDA
-			     runtime libs (cudart64_13 / cublas64_13 / cublasLt64_13) are NOT bundled
-			     — the consumer's installed CUDA Toolkit provides them. Mirrors the
-			     cuda (Linux) classifier pattern. Staged by CI before this profile runs. -->
-			<id>cuda-windows</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-compiler-plugin</artifactId>
-						<executions>
-						<execution>
-							<id>cuda-windows</id>
-							<phase>compile</phase>
-							<goals>
-								<goal>compile</goal>
-							</goals>
-							<configuration>
-								<excludes>
-									<exclude>module-info.java</exclude>
-								</excludes>
-								<compilerArgs>
-									<arg>-h</arg>
-									<arg>src/main/cpp</arg>
-								</compilerArgs>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_cuda</outputDirectory>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<artifactId>maven-resources-plugin</artifactId>
-						<executions>
-						<execution>
-							<id>copy-resources-cuda-windows</id>
-							<phase>process-classes</phase>
-							<goals>
-								<goal>copy-resources</goal>
-							</goals>
-							<configuration>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_cuda</outputDirectory>
-								<resources>
-									<resource>
-										<directory>
-											${basedir}/src/main/resources_windows_cuda/</directory>
-										<includes>
-											<include>**/*.*</include>
-										</includes>
-									</resource>
-								</resources>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-jar-plugin</artifactId>
-						<executions>
-							<execution>
-								<id>cuda-windows</id>
-								<phase>package</phase>
-								<goals>
-									<goal>jar</goal>
-								</goals>
-								<configuration>
-									<classifier>cuda13-windows-x86-64</classifier>
-									<classesDirectory>
-										${project.build.outputDirectory}_windows_cuda</classesDirectory>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-
-		<profile>
-			<!-- Windows Vulkan GPU natives (x86_64 only), shipped as the
-			     `vulkan-windows-x86-64` classifier JAR. Built with -DGGML_VULKAN=ON and
-			     routed by CMakeLists.txt to src/main/resources_windows_vulkan/. The Vulkan
-			     runtime (vulkan-1.dll) ships with the GPU driver — nothing is bundled.
-			     Vulkan is the most portable Windows GPU backend (NVIDIA/AMD/Intel).
-			     Staged by CI before this profile runs. -->
-			<id>vulkan-windows</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-compiler-plugin</artifactId>
-						<executions>
-						<execution>
-							<id>vulkan-windows</id>
-							<phase>compile</phase>
-							<goals>
-								<goal>compile</goal>
-							</goals>
-							<configuration>
-								<excludes>
-									<exclude>module-info.java</exclude>
-								</excludes>
-								<compilerArgs>
-									<arg>-h</arg>
-									<arg>src/main/cpp</arg>
-								</compilerArgs>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_vulkan</outputDirectory>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<artifactId>maven-resources-plugin</artifactId>
-						<executions>
-						<execution>
-							<id>copy-resources-vulkan-windows</id>
-							<phase>process-classes</phase>
-							<goals>
-								<goal>copy-resources</goal>
-							</goals>
-							<configuration>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_vulkan</outputDirectory>
-								<resources>
-									<resource>
-										<directory>
-											${basedir}/src/main/resources_windows_vulkan/</directory>
-										<includes>
-											<include>**/*.*</include>
-										</includes>
-									</resource>
-								</resources>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-jar-plugin</artifactId>
-						<executions>
-							<execution>
-								<id>vulkan-windows</id>
-								<phase>package</phase>
-								<goals>
-									<goal>jar</goal>
-								</goals>
-								<configuration>
-									<classifier>vulkan-windows-x86-64</classifier>
-									<classesDirectory>
-										${project.build.outputDirectory}_windows_vulkan</classesDirectory>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-
-		<profile>
-			<!-- Windows OpenCL GPU natives (x86_64 only), shipped as the
-			     `opencl-windows-x86-64` classifier JAR. Built with -DGGML_OPENCL=ON and
-			     routed by CMakeLists.txt to src/main/resources_windows_opencl/. The OpenCL
-			     ICD (System32\OpenCL.dll) ships with the GPU driver — nothing is bundled.
-			     NOTE: the GGML OpenCL backend is Adreno-tuned; on desktop GPUs CUDA/Vulkan
-			     are better supported. Staged by CI before this profile runs. -->
-			<id>opencl-windows</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-compiler-plugin</artifactId>
-						<executions>
-						<execution>
-							<id>opencl-windows</id>
-							<phase>compile</phase>
-							<goals>
-								<goal>compile</goal>
-							</goals>
-							<configuration>
-								<excludes>
-									<exclude>module-info.java</exclude>
-								</excludes>
-								<compilerArgs>
-									<arg>-h</arg>
-									<arg>src/main/cpp</arg>
-								</compilerArgs>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_opencl</outputDirectory>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<artifactId>maven-resources-plugin</artifactId>
-						<executions>
-						<execution>
-							<id>copy-resources-opencl-windows</id>
-							<phase>process-classes</phase>
-							<goals>
-								<goal>copy-resources</goal>
-							</goals>
-							<configuration>
-								<outputDirectory>
-									${project.build.outputDirectory}_windows_opencl</outputDirectory>
-								<resources>
-									<resource>
-										<directory>
-											${basedir}/src/main/resources_windows_opencl/</directory>
-										<includes>
-											<include>**/*.*</include>
-										</includes>
-									</resource>
-								</resources>
-							</configuration>
-						</execution>
-						</executions>
-					</plugin>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-jar-plugin</artifactId>
-						<executions>
-							<execution>
-								<id>opencl-windows</id>
-								<phase>package</phase>
-								<goals>
-									<goal>jar</goal>
-								</goals>
-								<configuration>
-									<classifier>opencl-windows-x86-64</classifier>
-									<classesDirectory>
-										${project.build.outputDirectory}_windows_opencl</classesDirectory>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-
-		<profile>
-			<id>vmlens</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>com.vmlens</groupId>
-						<artifactId>vmlens-maven-plugin</artifactId>
-						<configuration>
-							<!--
-								Run vmlens interleaving analysis over the whole `vmlens` test
-								package (smoke test + SessionStateInterleavingTest). The `com.vmlens:api`
-								test dependency lives in the main <dependencies> block.
-								Expand <includes> as more concurrency tests are added (the
-								streambuffer repo runs vmlens over its whole suite).
-							-->
-							<includes>
-								<include>**/vmlens/*.java</include>
-							</includes>
-						</configuration>
-						<executions>
-							<execution>
-								<id>vmlens-test</id>
-								<goals>
-									<goal>test</goal>
-								</goals>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-		<profile>
-			<id>jcstress</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.codehaus.mojo</groupId>
-						<artifactId>exec-maven-plugin</artifactId>
-						<executions>
-							<execution>
-								<id>jcstress</id>
-								<phase>test</phase>
-								<goals><goal>exec</goal></goals>
-								<configuration>
-									<executable>${java.home}/bin/java</executable>
-									<classpathScope>test</classpathScope>
-									<arguments>
-										<argument>-classpath</argument>
-										<classpath/>
-										<argument>org.openjdk.jcstress.Main</argument>
-										<argument>-v</argument>
-										<argument>-m</argument>
-										<argument>default</argument>
-									</arguments>
-								</configuration>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
-		<profile>
-			<!--
-				Builds the fat jar-with-dependencies uber JAR: the library classes, the
-				default-platform native libs from src/main/resources, and all runtime Java
-				dependencies in one drop-on-classpath JAR, runnable via the OpenAiCompatServer
-				Main-Class (set below) to start the OpenAI-compatible HTTP server. Off by
-				default; the CI `package` job activates it so the uber JAR rides along in the
-				`llama-jars` upload-artifact bundle. Documented in CLAUDE.md "Build Commands"
-				as `mvn -P assembly package`.
-			-->
-			<id>assembly</id>
-			<build>
-				<plugins>
-					<plugin>
-						<groupId>org.apache.maven.plugins</groupId>
-						<artifactId>maven-assembly-plugin</artifactId>
-						<configuration>
-							<descriptorRefs>
-								<descriptorRef>jar-with-dependencies</descriptorRef>
-							</descriptorRefs>
-							<archive>
-								<manifest>
-									<mainClass>net.ladenthin.llama.server.OpenAiCompatServer</mainClass>
-								</manifest>
-							</archive>
-						</configuration>
-						<executions>
-							<execution>
-								<id>build-fat-jar</id>
-								<phase>package</phase>
-								<goals>
-									<goal>single</goal>
-								</goals>
-							</execution>
-						</executions>
-					</plugin>
-				</plugins>
-			</build>
-		</profile>
 	</profiles>
 </project>