fix(ci): remove invalid sm_121 from cuda matrix, add GGML_BACKEND_DL to CPU build, and fix Windows console encoding

.github/workflows/build-cachyllama-all.yml

@@ -81,6 +81,7 @@ jobs:
   ubuntu-cpu:
     runs-on: ${{ matrix.os }}
     strategy:
+      fail-fast: false
       matrix:
         include:
           - build: 'x64'
@@ -106,6 +107,7 @@ jobs:
           cmake -B build -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
             -DGGML_CPU_ALL_VARIANTS=ON \
+            -DGGML_BACKEND_DL=ON \
             -DGGML_NATIVE=OFF \
             -DGGML_OPENMP=ON \
             -DCMAKE_INSTALL_RPATH='$ORIGIN' \
@@ -113,6 +115,14 @@ jobs:
             ${{ env.CMAKE_ARGS }}
           cmake --build build --config Release -j $(nproc)
 
+      - name: Verify CPU Binaries
+        run: |
+          cd CachyLLama/build/bin
+          export LD_LIBRARY_PATH="$(pwd):${LD_LIBRARY_PATH:-}"
+          ./llama-cli --help | head -n 10
+          ./llama-server --help | head -n 10
+          echo "CPU binary verification successful!"
+
       - name: Pack artifacts
         run: |
           cp CachyLLama/LICENSE CachyLLama/build/bin/
@@ -129,7 +139,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        sm: ${{ github.event_name == 'pull_request' && fromJSON('["89"]') || fromJSON('["75", "80", "86", "89", "90", "100", "120", "121"]') }}
+        sm: ${{ github.event_name == 'pull_request' && fromJSON('["89"]') || fromJSON('["75", "80", "86", "89", "90", "100", "120"]') }}
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4

.github/workflows/build-cachyllama-rocm.yml

@@ -582,10 +582,16 @@ jobs:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |
           TAG="${{ steps.tag.outputs.tag }}"
-          gh release create "$TAG" \
-            --title "CachyLLama $TAG (ROCm Nightly)" \
-            --notes-file "RELEASE_NOTES.md" \
-            ./release-packages/*.zip
+          if gh release view "$TAG" >/dev/null 2>&1; then
+            echo "Release $TAG already exists. Uploading/updating release assets with --clobber..."
+            gh release upload "$TAG" ./release-packages/*.zip --clobber
+          else
+            echo "Creating new release $TAG..."
+            gh release create "$TAG" \
+              --title "CachyLLama $TAG (ROCm Nightly)" \
+              --notes-file "RELEASE_NOTES.md" \
+              ./release-packages/*.zip
+          fi
 
       - name: Trigger Homebrew Tap Formula Update
         if: success()

scripts/patch_windows_compat.py

@@ -11,6 +11,9 @@ madvise(), POSIX mkdir) when building on MSVC or Clang on Windows.
 import sys
 from pathlib import Path
 
+if hasattr(sys.stdout, "reconfigure"):
+    sys.stdout.reconfigure(encoding="utf-8", errors="replace")
+
 
 def patch_file(path: Path, replacements: list[tuple[str, str]]) -> bool:
     if not path.exists():
@@ -28,7 +31,7 @@ def patch_file(path: Path, replacements: list[tuple[str, str]]) -> bool:
 
     if content != original:
         path.write_text(content, encoding="utf-8")
-        print(f"[✓] Patched {path}")
+        print(f"[+] Patched {path}")
         return True
     else:
         print(f"[*] No changes applied to {path}")
@@ -71,7 +74,7 @@ def main():
         ],
     )
 
-    print("[✓] CachyLLama Windows compatibility patch completed successfully.")
+    print("[+] CachyLLama Windows compatibility patch completed successfully.")
 
 
 if __name__ == "__main__":