Bump PyTorch pin to dev20250602

Found the commit hash from https://hud2.pytorch.org/hud/pytorch/pytorch/nightly/0 .

ghstack-source-id: 9840d13
ghstack-comment-id: 2936530681
Pull-Request-resolved: #11318

Author: swolchok

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-01f1cc44cbbfdf6307aa01b803a4ee22f9ade946
+64247892a0ca8ed045ad0b530eb87c3dd66590ea

install_requirements.py

@@ -71,7 +71,7 @@ def python_is_compatible():
 #
 # NOTE: If you're changing, make the corresponding change in .ci/docker/ci_commit_pins/pytorch.txt
 # by picking the hash from the same date in https://hud.pytorch.org/hud/pytorch/pytorch/nightly/
-NIGHTLY_VERSION = "dev20250422"
+NIGHTLY_VERSION = "dev20250602"
 
 
 def install_requirements(use_pytorch_nightly):
@@ -82,7 +82,7 @@ def install_requirements(use_pytorch_nightly):
         # been installed on CI before this step, so pip won't reinstall them
         f"torch==2.8.0.{NIGHTLY_VERSION}" if use_pytorch_nightly else "torch",
         (
-            f"torchvision==0.22.0.{NIGHTLY_VERSION}"
+            f"torchvision==0.23.0.{NIGHTLY_VERSION}"
             if use_pytorch_nightly
             else "torchvision"
         ),  # For testing.

runtime/core/portable_type/c10/c10/macros/Macros.h

@@ -241,7 +241,7 @@ using namespace c10::xpu;
 #ifdef __HIPCC__
 // Unlike CUDA, HIP requires a HIP header to be included for __host__ to work.
 // We do this #include here so that C10_HOST_DEVICE and friends will Just Work.
-// See https://github.com/ROCm-Developer-Tools/HIP/issues/441
+// See https://github.com/ROCm/hip/issues/441
 #include <hip/hip_runtime.h>
 #endif
 
@@ -286,7 +286,7 @@ constexpr uint32_t CUDA_THREADS_PER_BLOCK_FALLBACK = 256;
 #define C10_MIN_BLOCKS_PER_SM(threads_per_block, blocks_per_sm)        \
   ((((threads_per_block) * (blocks_per_sm) <= CUDA_MAX_THREADS_PER_SM) \
         ? (blocks_per_sm)                                              \
-        : ((CUDA_MAX_THREADS_PER_SM + (threads_per_block)-1) /         \
+        : ((CUDA_MAX_THREADS_PER_SM + (threads_per_block) - 1) /       \
            (threads_per_block))))
 // C10_LAUNCH_BOUNDS is analogous to __launch_bounds__
 #define C10_LAUNCH_BOUNDS_0 \

runtime/core/portable_type/c10/c10/util/BFloat16-inl.h

@@ -10,14 +10,11 @@ C10_CLANG_DIAGNOSTIC_PUSH()
 C10_CLANG_DIAGNOSTIC_IGNORE("-Wimplicit-int-float-conversion")
 #endif
 
-#if defined(SYCL_EXT_ONEAPI_BFLOAT16_MATH_FUNCTIONS)
 #if defined(CL_SYCL_LANGUAGE_VERSION)
 #include <CL/sycl.hpp> // for SYCL 1.2.1
-#else
+#elif defined(SYCL_LANGUAGE_VERSION)
 #include <sycl/sycl.hpp> // for SYCL 2020
 #endif
-#include <ext/oneapi/bfloat16.hpp>
-#endif
 
 namespace c10 {
 

runtime/core/portable_type/c10/c10/util/BFloat16.h

@@ -14,14 +14,11 @@
 #include <cuda_bf16.h>
 #endif
 
-#if defined(SYCL_EXT_ONEAPI_BFLOAT16_MATH_FUNCTIONS)
 #if defined(CL_SYCL_LANGUAGE_VERSION)
 #include <CL/sycl.hpp> // for SYCL 1.2.1
-#else
+#elif defined(SYCL_LANGUAGE_VERSION)
 #include <sycl/sycl.hpp> // for SYCL 2020
 #endif
-#include <ext/oneapi/bfloat16.hpp>
-#endif
 
 namespace c10 {