updated cpu build

Author: ekjotsinghmakhija

.github/workflows/build-cpu-release.yml

@@ -1,41 +1,41 @@
 name: Build Stable CPU Release (v0.3.16)
 
 on:
-  workflow_dispatch: # Allows you to manually trigger the build
+  workflow_dispatch:
 
 permissions:
-  contents: write # Needed to upload the release
+  contents: write
 
 jobs:
   build_cpu_wheel:
     name: Build CPU Wheel for Hugging Face
-    runs-on: ubuntu-latest # Matches Hugging Face's OS
+    runs-on: ubuntu-latest
 
     steps:
       - uses: actions/checkout@v4
         with:
-          # ⚡ TIME TRAVEL ⚡
-          # We explicitly fetch the latest stable v0.3.16 tag.
+          # ⚡ FIX: FETCH FROM UPSTREAM ⚡
+          # We grab the code from the original author (abetlen) because
+          # your fork might be missing the tags.
+          repository: abetlen/llama-cpp-python
           ref: v0.3.16
           submodules: recursive
 
       - name: Set up Python 3.11
         uses: actions/setup-python@v5
         with:
-          python-version: '3.11' # Matches your Dockerfile
+          python-version: '3.11'
 
       - name: Install Build Tools
         run: |
           pip install build wheel scikit-build-core cmake
 
       - name: ⚡ Compile Optimized CPU Wheel (v0.3.16) ⚡
         env:
-          # Force CPU-only build.
-          # We use "pthread" to fix the linker error you saw earlier.
+          # Force CPU-only build with pthread support
           CMAKE_ARGS: "-DLLAMA_BLAS=OFF -DLLAMA_CUBLAS=OFF -DLLAMA_METAL=OFF -DCMAKE_C_FLAGS='-pthread' -DCMAKE_CXX_FLAGS='-pthread'"
           FORCE_CMAKE: "1"
         run: |
-          # Build the wheel
           python -m build --wheel
 
       - name: Create Release