Fix Windows GPU CI: bump Jimver CUDA + switch Vulkan SDK action

First CI run on PR #276 (run 28327740376): default Ninja flip, MSVC
classifier, and the OpenCL job were green; CUDA and Vulkan failed on
toolchain setup.

- CUDA: `Jimver/cuda-toolkit@v0.2.24` errored "Version not available:
  13.0.0" (that tag predates CUDA 13.x). Bump to @v0.2.35 and request
  13.2.0 (matches the Linux pin; classifier stays cuda13-windows-x86-64).
  Also clears the Node 20 deprecation warning.
- Vulkan: humbletim/install-vulkan-sdk set VULKAN_SDK but laid out the
  SDK in a way CMake's FindVulkan could not read (missing Vulkan_LIBRARY
  / Vulkan_INCLUDE_DIR). Switch to jakoch/install-vulkan-sdk-action@v1.6.0
  (purpose-built, FindVulkan-compatible), vulkan_version 1.4.350.0.

Docs (CLAUDE.md, TODO.md) updated with the resolved toolchain versions
and the first-run results.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Deqf9xS6jz9t1idytVTaPV

Author: claude

.github/workflows/publish.yml

@@ -701,10 +701,10 @@ jobs:
         with:
           arch: x64
       - name: Install CUDA Toolkit
-        uses: Jimver/cuda-toolkit@v0.2.24
+        uses: Jimver/cuda-toolkit@v0.2.35
         id: cuda-toolkit
         with:
-          cuda: '13.0.0'
+          cuda: '13.2.0'
           method: 'network'
           sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust"]'
       - name: Install sccache (shared compiler cache)
@@ -751,9 +751,9 @@ jobs:
         with:
           arch: x64
       - name: Install Vulkan SDK
-        uses: humbletim/install-vulkan-sdk@v1.2
+        uses: jakoch/install-vulkan-sdk-action@v1.6.0
         with:
-          version: latest
+          vulkan_version: 1.4.350.0
           cache: true
       - name: Install sccache (shared compiler cache)
         if: env.USE_CACHE == 'true' && env.SCCACHE_WEBDAV_TOKEN != ''

CLAUDE.md

@@ -209,8 +209,9 @@ Wiring (mirrors the CUDA-Linux / OpenCL-Android classifier pattern):
    - `build-windows-x86_64` / `build-windows-x86` — **Ninja CPU**, artifacts `Windows-{arch}-libraries`
      → picked up by the `package` job's `pattern: "*-libraries"` into the **default** tree.
    - `build-windows-x86_64-msvc` / `build-windows-x86-msvc` — **MSVC CPU**, artifacts `Windows-{arch}-msvc`.
-   - `build-windows-x86_64-cuda` — `Jimver/cuda-toolkit` + `-DGGML_CUDA=ON`, artifact `Windows-x86_64-cuda`.
-   - `build-windows-x86_64-vulkan` — `humbletim/install-vulkan-sdk` + `-DGGML_VULKAN=ON`, artifact
+   - `build-windows-x86_64-cuda` — `Jimver/cuda-toolkit@v0.2.35` (CUDA `13.2.0`) + `-DGGML_CUDA=ON`,
+     artifact `Windows-x86_64-cuda`.
+   - `build-windows-x86_64-vulkan` — `jakoch/install-vulkan-sdk-action` + `-DGGML_VULKAN=ON`, artifact
      `Windows-x86_64-vulkan`.
    - `build-windows-x86_64-opencl` — `build_opencl_windows.bat -DGGML_OPENCL=ON -DGGML_OPENCL_EMBED_KERNELS=ON`,
      artifact `Windows-x86_64-opencl`.
@@ -224,11 +225,13 @@ Wiring (mirrors the CUDA-Linux / OpenCL-Android classifier pattern):
 
 `src/main/resources_windows_{msvc,cuda,vulkan,opencl}/` are git-ignored (staged by CI, never committed).
 
-**First-run unknowns (no GPU/Windows CI tested locally — validate on the first CI run):** the exact
-`Jimver/cuda-toolkit` CUDA version that resolves on Windows (the job requests `13.0.0`; if Jimver has
-no 13.x the classifier would need to drop to `cuda12-…`), the `humbletim/install-vulkan-sdk` action
-inputs, and the `build_opencl_windows.bat` ICD-loader staging are all unproven on the runner image and
-may need adjustment. Because all five Windows build jobs are in the `package`/publish `needs:` graph, a
+**First CI run (PR #276, run 28327740376):** the default Ninja CPU flip, the MSVC classifier, and the
+**OpenCL** job were green on the first try. Two GPU jobs needed a toolchain fix: **CUDA** failed with
+`Version not available: 13.0.0` because the pinned `Jimver/cuda-toolkit@v0.2.24` predated CUDA 13.x →
+bumped to `@v0.2.35` + `13.2.0` (matches the Linux pin, classifier stays `cuda13-…`); **Vulkan** failed
+`find_package(Vulkan)` because `humbletim/install-vulkan-sdk` set `VULKAN_SDK` but laid the SDK out in a
+way CMake's `FindVulkan` couldn't read → switched to `jakoch/install-vulkan-sdk-action` (purpose-built,
+FindVulkan-compatible). Because all five Windows build jobs are in the `package`/publish `needs:` graph, a
 GPU-toolchain failure blocks packaging — the same release-gating policy the Linux-CUDA / Android-OpenCL
 jobs already follow.
 

TODO.md

@@ -197,8 +197,8 @@ Multi-Config + MSVC.
   **MSVC** (artifacts `Windows-{arch}-msvc`). `test-java-windows-x86_64` (default/Ninja) and
   `test-java-windows-x86_64-msvc` both load the DLL via JNI and run the full model-backed suite.
 - **GPU build jobs (x86_64, Ninja, build + `ctest` only — runners have no GPU):**
-  `build-windows-x86_64-cuda` (`Jimver/cuda-toolkit` + `-DGGML_CUDA=ON`),
-  `build-windows-x86_64-vulkan` (`humbletim/install-vulkan-sdk` + `-DGGML_VULKAN=ON`),
+  `build-windows-x86_64-cuda` (`Jimver/cuda-toolkit@v0.2.35` CUDA `13.2.0` + `-DGGML_CUDA=ON`),
+  `build-windows-x86_64-vulkan` (`jakoch/install-vulkan-sdk-action` + `-DGGML_VULKAN=ON`),
   `build-windows-x86_64-opencl` (`build_opencl_windows.bat` stages the ICD loader + `-DGGML_OPENCL=ON`).
 - **`CMakeLists.txt`** — OS-aware backend routing (CUDA/OpenCL → Windows trees, new Vulkan branch).
 - **`.github/build.bat`** — also wraps nvcc with sccache for CUDA builds.
@@ -210,12 +210,12 @@ Multi-Config + MSVC.
   four profiles; all five Windows build jobs are in the `package` `needs:` graph.
 - Docs: `README.md` classifier table + `CLAUDE.md` "Windows native classifiers" section.
 
-**Verification — PENDING.** GPU/Windows builds could not be exercised locally (no Windows host, no
-GPU). The first CI run must confirm: the `Jimver/cuda-toolkit` CUDA version resolves on Windows (the
-job requests `13.0.0`; drop the classifier to `cuda12-…` if Jimver has no 13.x), the
-`humbletim/install-vulkan-sdk` inputs are correct, and `build_opencl_windows.bat`'s ICD-loader staging
-works on the runner image. The default Ninja CPU flip should be green immediately (it reuses the
-already-proven Ninja jobs, just renamed).
+**Verification — first CI run done (PR #276, run 28327740376).** Green on the first try: default Ninja
+CPU flip (x64+x86), MSVC classifier (x64+x86), and the **OpenCL** GPU job (`build_opencl_windows.bat`
+ICD staging works). Two GPU jobs were fixed after the first run: **CUDA** (`Version not available:
+13.0.0` → bumped `Jimver/cuda-toolkit` `v0.2.24`→`v0.2.35` + `13.2.0`) and **Vulkan**
+(`find_package(Vulkan)` couldn't read the `humbletim` SDK layout → switched to
+`jakoch/install-vulkan-sdk-action`). Re-run pending to confirm both fixes.
 
 **Optional follow-up:** smoke-test that each *published* classifier JAR loads its DLL on a clean
 Windows host with the matching GPU driver/toolkit installed.