Add windows 12.9 domain builds (#6838)

pytorch/pytorch#155196

Add Windows 12.9 domain builds (torchaudio/torchvision) 

Also removes the unnecessary cuda-aarch64 entry, it is a duplicate
entry.

cc @atalman

Author: tinglvv

tools/scripts/generate_binary_build_matrix.py

@@ -176,9 +176,7 @@ def initialize_globals(channel: str, build_python_only: bool) -> None:
         },
         CPU: "pytorch/manylinux2_28-builder:cpu",
         XPU: "pytorch/manylinux2_28-builder:xpu",
-        # TODO: Migrate CUDA_AARCH64 image to manylinux2_28_aarch64-builder:cuda12.4
         CPU_AARCH64: "pytorch/manylinux2_28_aarch64-builder:cpu-aarch64",
-        CUDA_AARCH64: "pytorch/manylinuxaarch64-builder:cuda12.6",
     }
     LIBTORCH_CONTAINER_IMAGES = {
         **{
@@ -445,9 +443,6 @@ def generate_wheels_matrix(
             upload_to_base_bucket = "no"
             if os in (LINUX, WINDOWS):
                 arches += CUDA_ARCHES
-                # todo: remove once windows cuda 12.9 binaries are available
-                if channel == NIGHTLY and os != LINUX:
-                    arches.remove("12.9")
             elif os == LINUX_AARCH64:
                 arches += CUDA_AARCH64_ARCHES
 

tools/tests/assets/build_matrix_windows_wheel_cuda.json

@@ -45,6 +45,21 @@
       "stable_version": "2.7.1",
       "use_split_build": false
     },
+    {
+      "python_version": "3.9",
+      "gpu_arch_type": "cuda",
+      "gpu_arch_version": "12.9",
+      "desired_cuda": "cu129",
+      "container_image": "pytorch/manylinux2_28-builder:cuda12.9",
+      "package_type": "wheel",
+      "build_name": "wheel-py3_9-cuda12_9",
+      "validation_runner": "windows.g4dn.xlarge",
+      "installation": "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129",
+      "channel": "nightly",
+      "upload_to_base_bucket": "no",
+      "stable_version": "2.7.1",
+      "use_split_build": false
+    },
     {
       "python_version": "3.9",
       "gpu_arch_type": "xpu",
@@ -105,6 +120,21 @@
       "stable_version": "2.7.1",
       "use_split_build": false
     },
+    {
+      "python_version": "3.10",
+      "gpu_arch_type": "cuda",
+      "gpu_arch_version": "12.9",
+      "desired_cuda": "cu129",
+      "container_image": "pytorch/manylinux2_28-builder:cuda12.9",
+      "package_type": "wheel",
+      "build_name": "wheel-py3_10-cuda12_9",
+      "validation_runner": "windows.g4dn.xlarge",
+      "installation": "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129",
+      "channel": "nightly",
+      "upload_to_base_bucket": "no",
+      "stable_version": "2.7.1",
+      "use_split_build": false
+    },
     {
       "python_version": "3.10",
       "gpu_arch_type": "xpu",
@@ -165,6 +195,21 @@
       "stable_version": "2.7.1",
       "use_split_build": false
     },
+    {
+      "python_version": "3.11",
+      "gpu_arch_type": "cuda",
+      "gpu_arch_version": "12.9",
+      "desired_cuda": "cu129",
+      "container_image": "pytorch/manylinux2_28-builder:cuda12.9",
+      "package_type": "wheel",
+      "build_name": "wheel-py3_11-cuda12_9",
+      "validation_runner": "windows.g4dn.xlarge",
+      "installation": "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129",
+      "channel": "nightly",
+      "upload_to_base_bucket": "no",
+      "stable_version": "2.7.1",
+      "use_split_build": false
+    },
     {
       "python_version": "3.11",
       "gpu_arch_type": "xpu",
@@ -225,6 +270,21 @@
       "stable_version": "2.7.1",
       "use_split_build": false
     },
+    {
+      "python_version": "3.12",
+      "gpu_arch_type": "cuda",
+      "gpu_arch_version": "12.9",
+      "desired_cuda": "cu129",
+      "container_image": "pytorch/manylinux2_28-builder:cuda12.9",
+      "package_type": "wheel",
+      "build_name": "wheel-py3_12-cuda12_9",
+      "validation_runner": "windows.g4dn.xlarge",
+      "installation": "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129",
+      "channel": "nightly",
+      "upload_to_base_bucket": "no",
+      "stable_version": "2.7.1",
+      "use_split_build": false
+    },
     {
       "python_version": "3.12",
       "gpu_arch_type": "xpu",
@@ -285,6 +345,21 @@
       "stable_version": "2.7.1",
       "use_split_build": false
     },
+    {
+      "python_version": "3.13",
+      "gpu_arch_type": "cuda",
+      "gpu_arch_version": "12.9",
+      "desired_cuda": "cu129",
+      "container_image": "pytorch/manylinux2_28-builder:cuda12.9",
+      "package_type": "wheel",
+      "build_name": "wheel-py3_13-cuda12_9",
+      "validation_runner": "windows.g4dn.xlarge",
+      "installation": "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129",
+      "channel": "nightly",
+      "upload_to_base_bucket": "no",
+      "stable_version": "2.7.1",
+      "use_split_build": false
+    },
     {
       "python_version": "3.13",
       "gpu_arch_type": "xpu",
@@ -345,6 +420,21 @@
       "stable_version": "2.7.1",
       "use_split_build": false
     },
+    {
+      "python_version": "3.13t",
+      "gpu_arch_type": "cuda",
+      "gpu_arch_version": "12.9",
+      "desired_cuda": "cu129",
+      "container_image": "pytorch/manylinux2_28-builder:cuda12.9",
+      "package_type": "wheel",
+      "build_name": "wheel-py3_13t-cuda12_9",
+      "validation_runner": "windows.g4dn.xlarge",
+      "installation": "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129",
+      "channel": "nightly",
+      "upload_to_base_bucket": "no",
+      "stable_version": "2.7.1",
+      "use_split_build": false
+    },
     {
       "python_version": "3.13t",
       "gpu_arch_type": "xpu",