[XPU] xpu kernel release workflow (#33679)

scripts/update_kernel_whl_index.py

@@ -11,10 +11,12 @@ DEFAULT_CUDA_VERSION = "130"
 
 
 def check_wheel_cuda_version(path_name, target_cuda_version):
-    # Skip non-CUDA backend wheels (rocm, musa, ...). Their +<backend><ver>
-    # local-version tags don't match the CUDA wheel regex below, and they are
-    # published by the dedicated release-rocm*/release-musa* jobs.
-    if re.search(r"\+(rocm|musa)", path_name):
+    # Skip non-CUDA backend wheels. ROCm/MUSA encode the backend in the
+    # local-version tag (for example +rocm720), while XPU uses a dedicated
+    # package name (sglang_kernel_xpu-*).
+    if re.search(r"\+(rocm|musa)", path_name) or path_name.startswith(
+        "sglang_kernel_xpu-"
+    ):
         return False
 
     # For other CUDA versions, the wheel path name will contain the cuda version suffix, e.g. sglang_kernel-0.4.0+cu130-cp310-abi3-manylinux2014_x86_64.whl
@@ -48,8 +50,14 @@ def update_wheel_index(cuda_version=DEFAULT_CUDA_VERSION, rocm_version=None):
             f.write(f'<a href="{full_url}">{path.name}</a><br>\n')
 
 
-def _update_non_cuda_wheel_index(backend, version):
-    index_dir = pathlib.Path(f"sgl-whl/{backend}{version}/sglang-kernel")
+def _update_non_cuda_wheel_index(
+    backend,
+    version=None,
+    package_name="sglang_kernel",
+    index_package_name="sglang-kernel",
+):
+    backend_dir = f"{backend}{version or ''}"
+    index_dir = pathlib.Path(f"sgl-whl/{backend_dir}/{index_package_name}")
     index_dir.mkdir(exist_ok=True, parents=True)
     base_url = "https://github.com/sgl-project/whl/releases/download"
 
@@ -60,7 +68,7 @@ def _update_non_cuda_wheel_index(backend, version):
         with open(path, "rb") as f:
             sha256 = hashlib.sha256(f.read()).hexdigest()
         ver = re.findall(
-            rf"sglang_kernel-([0-9.]+(?:\.post[0-9]+)?)(?:\+{backend}[0-9]+)?-",
+            rf"{re.escape(package_name)}-([0-9.]+(?:\.post[0-9]+)?)(?:\+{backend}[0-9]+)?-",
             path.name,
         )[0]
         full_url = f"{base_url}/v{ver}/{path.name}#sha256={sha256}"
@@ -68,6 +76,14 @@ def _update_non_cuda_wheel_index(backend, version):
             f.write(f'<a href="{full_url}">{path.name}</a><br>\n')
 
 
+def update_wheel_index_xpu():
+    _update_non_cuda_wheel_index(
+        "xpu",
+        package_name="sglang_kernel_xpu",
+        index_package_name="sglang-kernel-xpu",
+    )
+
+
 def update_wheel_index_rocm(rocm_version):
     _update_non_cuda_wheel_index("rocm", rocm_version)
 
@@ -81,8 +97,11 @@ def main():
     parser.add_argument("--cuda", type=str, default=DEFAULT_CUDA_VERSION)
     parser.add_argument("--rocm", type=str, default=None)
     parser.add_argument("--musa", type=str, default=None)
+    parser.add_argument("--xpu", action="store_true")
     args = parser.parse_args()
-    if args.musa is not None:
+    if args.xpu:
+        update_wheel_index_xpu()
+    elif args.musa is not None:
         update_wheel_index_musa(args.musa)
     elif args.rocm is not None:
         update_wheel_index_rocm(args.rocm)