diff --git a/.ci/docker/ci_commit_pins/pytorch.txt b/.ci/docker/ci_commit_pins/pytorch.txt
index 401a0594d98..7a1fe71d429 100644
--- a/.ci/docker/ci_commit_pins/pytorch.txt
+++ b/.ci/docker/ci_commit_pins/pytorch.txt
@@ -1 +1 @@
-release/2.13
+69985f4c2d95401063ec5739934ba09d9aae3792
diff --git a/backends/vulkan/test/op_tests/CMakeLists.txt b/backends/vulkan/test/op_tests/CMakeLists.txt
index 0f8456accf5..59a38b1ce50 100644
--- a/backends/vulkan/test/op_tests/CMakeLists.txt
+++ b/backends/vulkan/test/op_tests/CMakeLists.txt
@@ -72,6 +72,7 @@ function(vulkan_op_test test_name test_src)
   set(extra_deps ${ARGN})
 
   add_executable(${test_name} ${test_src})
+  target_compile_features(${test_name} PRIVATE cxx_std_20)
   target_include_directories(${test_name} PRIVATE ${COMMON_INCLUDES})
   target_link_libraries(
     ${test_name}
@@ -89,6 +90,7 @@ endfunction()
 
 if(TARGET vulkan_backend AND LIB_TORCH)
   add_library(test_utils ${CMAKE_CURRENT_SOURCE_DIR}/test_utils.cpp)
+  target_compile_features(test_utils PRIVATE cxx_std_20)
   target_include_directories(test_utils PRIVATE ${COMMON_INCLUDES})
   target_link_libraries(
     test_utils PRIVATE vulkan_backend ${LIB_TORCH} ${LIB_TORCH_CPU}
diff --git a/backends/xnnpack/operators/quant_params.py b/backends/xnnpack/operators/quant_params.py
index 6a3acd0df39..224d6b92431 100644
--- a/backends/xnnpack/operators/quant_params.py
+++ b/backends/xnnpack/operators/quant_params.py
@@ -163,6 +163,26 @@ def __str__(self) -> str:
             f")"
         )
 
+    def matches(self, other: QuantParams) -> bool:
+        def values_match(lhs, rhs) -> bool:
+            if isinstance(lhs, torch.Tensor) and isinstance(rhs, torch.Tensor):
+                return torch.equal(lhs, rhs)
+            return bool(lhs == rhs)
+
+        return (
+            self.per_channel == other.per_channel
+            and self.per_channel_group == other.per_channel_group
+            and values_match(self.scale, other.scale)
+            and values_match(self.zp, other.zp)
+            and self.axis == other.axis
+            and self.dtype == other.dtype
+            and self.qmin == other.qmin
+            and self.qmax == other.qmax
+            and self.is_dynamic == other.is_dynamic
+            and self.num_nonbatch_dims == other.num_nonbatch_dims
+            and self.group_size == other.group_size
+        )
+
     def quantize_tensor(self, tensor: torch.Tensor) -> torch.Tensor:
         # Do nothing if already quantized by the Quantizer
         if tensor.dtype == self.dtype:
diff --git a/backends/xnnpack/partition/config/generic_node_configs.py b/backends/xnnpack/partition/config/generic_node_configs.py
index c7a3be5f65f..623d1a298de 100644
--- a/backends/xnnpack/partition/config/generic_node_configs.py
+++ b/backends/xnnpack/partition/config/generic_node_configs.py
@@ -12,6 +12,7 @@
 
 import numpy as np
 import torch
+from executorch.backends.xnnpack.operators.quant_params import QuantParams
 from executorch.backends.xnnpack.partition.config.xnnpack_config import (
     ConfigPrecisionType,
     XNNPartitionerConfig,
@@ -418,6 +419,21 @@ def check_constraints(self, node: torch.fx.Node, ep: ExportedProgram) -> bool:
         if not self.check_common_constraints(node, ep):
             return False
 
+        input_node = get_input_node(node, 0)
+        output_node = next(iter(node.users), None)
+        if is_dequant(input_node) and output_node is not None and is_quant(output_node):
+            input_quant_params = QuantParams.from_q_dq_node(input_node, ep)
+            output_quant_params = QuantParams.from_q_dq_node(output_node, ep)
+            if not input_quant_params.matches(output_quant_params):
+                why(
+                    node,
+                    reason=(
+                        "XNNPACK static reshape requires matching input and "
+                        "output quantization parameters."
+                    ),
+                )
+                return False
+
         new_shape = node.args[1]
 
         # Check for symbolic dims. They aren't lowerable to XNNPACK currently.
diff --git a/backends/xnnpack/test/test_xnnpack_partitioner.py b/backends/xnnpack/test/test_xnnpack_partitioner.py
index 894fab4098f..36f317014f7 100644
--- a/backends/xnnpack/test/test_xnnpack_partitioner.py
+++ b/backends/xnnpack/test/test_xnnpack_partitioner.py
@@ -12,11 +12,17 @@
 import torch.nn.functional as F
 
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
-from executorch.exir import to_edge, to_edge_transform_and_lower
+from executorch.backends.xnnpack.quantizer.xnnpack_quantizer import (
+    get_symmetric_quantization_config,
+    XNNPACKQuantizer,
+)
+from executorch.backends.xnnpack.utils.configs import get_transform_passes
+from executorch.exir import EdgeCompileConfig, to_edge, to_edge_transform_and_lower
 from executorch.extension.pybindings.portable_lib import (
     _load_for_executorch_from_buffer,
 )
 from torch.export import export
+from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
 
 
 class TestXnnpackPartitioner(unittest.TestCase):
@@ -88,6 +94,62 @@ def test_no_warning_for_to_edge_transform_and_lower_workflow(self):
         log_contents = log_capture_string.getvalue()
         self.assertNotIn("DEPRECATION WARNING", log_contents)
 
+    def test_quantized_view_with_mismatched_qparams_stays_portable(self):
+        class ConvPoolFlattenLinear(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+                self.conv = torch.nn.Conv2d(3, 4, 3)
+                self.linear = torch.nn.Linear(4, 2)
+
+            def forward(self, x):
+                x = self.conv(x)
+                x = F.adaptive_avg_pool2d(x, (1, 1))
+                return self.linear(torch.flatten(x, 1))
+
+        inputs = (torch.randn(1, 3, 8, 8),)
+        exported = export(ConvPoolFlattenLinear().eval(), inputs, strict=False)
+        quantizer = XNNPACKQuantizer().set_global(
+            get_symmetric_quantization_config(is_per_channel=False)
+        )
+        prepared = prepare_pt2e(exported.module(), quantizer)
+        prepared(*inputs)
+        converted = convert_pt2e(prepared)
+
+        output_quant = next(
+            node
+            for node in converted.graph.nodes
+            if "quantize_per_tensor" in str(node.target)
+            and any("flatten" in str(inp.target) for inp in node.all_input_nodes)
+        )
+        output_scale = output_quant.args[1] * 2
+        output_quant.args = (
+            output_quant.args[0],
+            output_scale,
+            *output_quant.args[2:],
+        )
+        output_dequant = next(iter(output_quant.users))
+        output_dequant.args = (
+            output_dequant.args[0],
+            output_scale,
+            *output_dequant.args[2:],
+        )
+        converted.recompile()
+
+        lowered = to_edge_transform_and_lower(
+            export(converted, inputs, strict=False),
+            partitioner=[XnnpackPartitioner()],
+            transform_passes=get_transform_passes(),
+            compile_config=EdgeCompileConfig(
+                _check_ir_validity=False, _skip_dim_order=True
+            ),
+        )
+        self.assertTrue(
+            any(
+                "view_copy" in str(node.target)
+                for node in lowered.exported_program().graph.nodes
+            )
+        )
+
     def test_multi_method_partitioning_with_shared_weights(self):
         """
         Test that multi-method models with shared weights are correctly partitioned.
diff --git a/runtime/core/portable_type/c10/c10/util/complex.h b/runtime/core/portable_type/c10/c10/util/complex.h
index 4e699684bc3..be99e270aa1 100644
--- a/runtime/core/portable_type/c10/c10/util/complex.h
+++ b/runtime/core/portable_type/c10/c10/util/complex.h
@@ -31,19 +31,11 @@ C10_HOST_DEVICE T abs(const c10::complex<T>& z) {
 #endif
 }
 
-#if defined(USE_ROCM)
-#define ROCm_Bug(x)
-#else
-#define ROCm_Bug(x) x
-#endif
-
 template <typename T>
 C10_HOST_DEVICE T arg(const c10::complex<T>& z) {
-  return ROCm_Bug(std)::atan2(std::imag(z), std::real(z));
+  return std::atan2(std::imag(z), std::real(z));
 }
 
-#undef ROCm_Bug
-
 template <typename T>
 constexpr T norm(const c10::complex<T>& z) {
   return z.real() * z.real() + z.imag() * z.imag();
diff --git a/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h b/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h
index da297241449..8ae5cde4f02 100644
--- a/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h
+++ b/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h
@@ -400,6 +400,7 @@ constexpr inline bool isShiftedUInt(uint64_t x) {
       N + S <= 64, "isShiftedUInt<N, S> with N + S > 64 is too wide.");
   // Per the two static_asserts above, S must be strictly less than 64.  So
   // 1 << S is not undefined behavior.
+  // NOLINTNEXTLINE(bugprone-chained-comparison)
   return isUInt<N + S>(x) && (x % (UINT64_C(1) << S) == 0);
 }
 
diff --git a/runtime/core/portable_type/c10/c10/util/overflows.h b/runtime/core/portable_type/c10/c10/util/overflows.h
index 183a2f62a32..348ee5ffa40 100644
--- a/runtime/core/portable_type/c10/c10/util/overflows.h
+++ b/runtime/core/portable_type/c10/c10/util/overflows.h
@@ -61,13 +61,28 @@ template <typename To, typename From>
 std::enable_if_t<std::is_floating_point_v<From>, bool> overflows(
     From f,
     bool strict_unsigned [[maybe_unused]] = false) {
-  using limit = std::numeric_limits<typename scalar_value_type<To>::type>;
+  using ToScalar = typename scalar_value_type<To>::type;
+  using limit = std::numeric_limits<ToScalar>;
   if (limit::has_infinity && std::isinf(static_cast<double>(f))) {
     return false;
   }
   if (!limit::has_quiet_NaN && (f != f)) {
     return true;
   }
+  if constexpr (std::is_integral_v<ToScalar>) {
+    // limit::max() for wide integer types is NOT exactly representable in
+    // floating point (e.g. int64 max = 2^63-1 rounds up to 2^63), so `f >
+    // limit::max()` lets a just-out-of-range value like 2^63 slip through and
+    // then become INT64_MIN via static_cast. Compare against the
+    // exactly-representable upper bound max()+1 == 2^digits instead. lowest()
+    // is 0 or a negated power of two, so it stays exact. (digits-1 keeps the
+    // shift < 64 for the uint64 case; the *2 recovers 2^digits without a 1<<64
+    // overflow.)
+    constexpr int digits = limit::digits;
+    constexpr From upper =
+        static_cast<From>(uint64_t{1} << (digits - 1)) * From{2};
+    return f < static_cast<From>(limit::lowest()) || f >= upper;
+  }
   return f < limit::lowest() || f > limit::max();
 }
 
diff --git a/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h b/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h
index cef99df3f56..08c4e9f1f84 100644
--- a/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h
+++ b/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h
@@ -123,6 +123,15 @@
 #define C10_HAS_CPP_ATTRIBUTE(x) (0)
 #endif
 
+/// Bind a returned reference/pointer's lifetime to a parameter (or *this) so
+/// Clang can warn when it would dangle. Expands to nothing on compilers that
+/// lack the attribute (e.g. non-clang, older nvcc).
+#if C10_HAS_CPP_ATTRIBUTE(clang::lifetimebound)
+#define C10_LIFETIMEBOUND [[clang::lifetimebound]]
+#else
+#define C10_LIFETIMEBOUND
+#endif
+
 #ifndef FBCODE_CAFFE2
 /// DEPRECATED: Warn if a type or return value is discarded.
 #define C10_NODISCARD [[nodiscard]]
diff --git a/runtime/core/portable_type/c10/torch/headeronly/util/Half.h b/runtime/core/portable_type/c10/torch/headeronly/util/Half.h
index e5aa622656c..401472357ec 100644
--- a/runtime/core/portable_type/c10/torch/headeronly/util/Half.h
+++ b/runtime/core/portable_type/c10/torch/headeronly/util/Half.h
@@ -213,7 +213,7 @@ C10_HOST_DEVICE inline float fp16_ieee_to_fp32_value(uint16_t h) {
    * Now, remember that denormalized half-precision numbers are represented as:
    *    FP16 = mantissa * 2**(-24).
    * The trick is to construct a normalized single-precision number with the
-   * same mantissa and thehalf-precision input and with an exponent which would
+   * same mantissa and the half-precision input and with an exponent which would
    * scale the corresponding mantissa bits to 2**(-24). A normalized
    * single-precision floating-point number is represented as: FP32 = (1 +
    * mantissa * 2**(-23)) * 2**(exponent - 127) Therefore, when the biased
diff --git a/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h b/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h
index c33a286bc5b..8e897957fee 100644
--- a/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h
+++ b/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h
@@ -14,20 +14,6 @@ C10_CLANG_DIAGNOSTIC_IGNORE("-Wimplicit-int-float-conversion")
 
 namespace c10 {
 
-/// Returns false since we cannot have x < 0 if x is unsigned.
-template <typename T>
-inline constexpr bool is_negative(
-    const T& /*x*/,
-    std::true_type /*is_unsigned*/) {
-  return false;
-}
-
-/// Returns true if a signed variable x < 0
-template <typename T>
-inline constexpr bool is_negative(const T& x, std::false_type /*is_unsigned*/) {
-  return x < T(0);
-}
-
 /// Returns true if x < 0
 /// NOTE: Will fail on an unsigned custom type
 ///       For the most part it's possible to fix this if
@@ -35,19 +21,12 @@ inline constexpr bool is_negative(const T& x, std::false_type /*is_unsigned*/) {
 ///       However, notably, c10::Half does not :-(
 template <typename T>
 inline constexpr bool is_negative(const T& x) {
-  return is_negative(x, std::is_unsigned<T>());
-}
-
-/// Returns the sign of an unsigned variable x as 0, 1
-template <typename T>
-inline constexpr int signum(const T& x, std::true_type /*is_unsigned*/) {
-  return T(0) < x;
-}
-
-/// Returns the sign of a signed variable x as -1, 0, 1
-template <typename T>
-inline constexpr int signum(const T& x, std::false_type /*is_unsigned*/) {
-  return (T(0) < x) - (x < T(0));
+  if constexpr (std::is_unsigned_v<T>) {
+    // An unsigned value can never be less than zero.
+    return false;
+  } else {
+    return x < T(0);
+  }
 }
 
 /// Returns the sign of x as -1, 0, 1
@@ -57,7 +36,11 @@ inline constexpr int signum(const T& x, std::false_type /*is_unsigned*/) {
 ///       However, notably, c10::Half does not :-(
 template <typename T>
 inline constexpr int signum(const T& x) {
-  return signum(x, std::is_unsigned<T>());
+  if constexpr (std::is_unsigned_v<T>) {
+    return T(0) < x;
+  } else {
+    return (T(0) < x) - (x < T(0));
+  }
 }
 
 /// Returns true if a and b are not both negative
@@ -86,53 +69,22 @@ inline constexpr bool greater_than_max(const T& x) {
 #pragma GCC diagnostic pop
 #endif
 
-/// Returns true if x < lowest(Limit). Standard comparison
-template <typename Limit, typename T>
-inline constexpr bool less_than_lowest(
-    const T& x,
-    std::false_type /*limit_is_unsigned*/,
-    std::false_type /*x_is_unsigned*/) {
-  return x < std::numeric_limits<Limit>::lowest();
-}
-
-/// Returns false since all the limit is signed and therefore includes
-/// negative values but x cannot be negative because it is unsigned
-template <typename Limit, typename T>
-inline constexpr bool less_than_lowest(
-    const T& /*x*/,
-    std::false_type /*limit_is_unsigned*/,
-    std::true_type /*x_is_unsigned*/) {
-  return false;
-}
-
-/// Returns true if x < 0, where 0 is constructed from T.
-/// Limit is not signed, so its lower value is zero
-template <typename Limit, typename T>
-inline constexpr bool less_than_lowest(
-    const T& x,
-    std::true_type /*limit_is_unsigned*/,
-    std::false_type /*x_is_unsigned*/) {
-  return x < T(0);
-}
-
-/// Returns false sign both types are unsigned
-template <typename Limit, typename T>
-inline constexpr bool less_than_lowest(
-    const T& /*x*/,
-    std::true_type /*limit_is_unsigned*/,
-    std::true_type /*x_is_unsigned*/) {
-  return false;
-}
-
-/// Returns true if x is less than the lowest value of type T
+/// Returns true if x is less than the lowest value of type Limit
 /// NOTE: Will fail on an unsigned custom type
 ///       For the most part it's possible to fix this if
 ///       the custom type has a constexpr constructor.
 ///       However, notably, c10::Half does not :
 template <typename Limit, typename T>
 inline constexpr bool less_than_lowest(const T& x) {
-  return less_than_lowest<Limit>(
-      x, std::is_unsigned<Limit>(), std::is_unsigned<T>());
+  if constexpr (std::is_unsigned_v<T>) {
+    // x is unsigned, so it can never be below the lowest value of any type.
+    return false;
+  } else if constexpr (std::is_unsigned_v<Limit>) {
+    // Limit is unsigned, so its lowest value is zero.
+    return x < T(0);
+  } else {
+    return x < std::numeric_limits<Limit>::lowest();
+  }
 }
 
 } // namespace c10
diff --git a/runtime/core/portable_type/c10/torch/headeronly/util/complex.h b/runtime/core/portable_type/c10/torch/headeronly/util/complex.h
index 733a22d5dbb..41430766952 100644
--- a/runtime/core/portable_type/c10/torch/headeronly/util/complex.h
+++ b/runtime/core/portable_type/c10/torch/headeronly/util/complex.h
@@ -3,6 +3,7 @@
 #include <complex>
 
 #include <torch/headeronly/macros/Macros.h>
+#include <torch/headeronly/util/BFloat16.h>
 #include <torch/headeronly/util/Half.h>
 
 #if defined(__CUDACC__) || defined(__HIPCC__)
@@ -588,6 +589,60 @@ struct alignas(4) complex<Half> {
   }
 };
 
+template <>
+struct alignas(4) complex<BFloat16> {
+  BFloat16 real_;
+  BFloat16 imag_;
+
+  // Constructors
+  complex() = default;
+  // BFloat16 constructor is not constexpr so the following constructor can't
+  // be constexpr
+  C10_HOST_DEVICE explicit inline complex(
+      const BFloat16& real,
+      const BFloat16& imag)
+      : real_(real), imag_(imag) {}
+  C10_HOST_DEVICE inline complex(const c10::complex<float>& value)
+      : real_(value.real()), imag_(value.imag()) {}
+
+  // Conversion operator
+  inline C10_HOST_DEVICE operator c10::complex<float>() const {
+    return {real_, imag_};
+  }
+
+  constexpr C10_HOST_DEVICE BFloat16 real() const {
+    return real_;
+  }
+  constexpr C10_HOST_DEVICE BFloat16 imag() const {
+    return imag_;
+  }
+
+  C10_HOST_DEVICE complex<BFloat16>& operator+=(
+      const complex<BFloat16>& other) {
+    real_ = static_cast<float>(real_) + static_cast<float>(other.real_);
+    imag_ = static_cast<float>(imag_) + static_cast<float>(other.imag_);
+    return *this;
+  }
+
+  C10_HOST_DEVICE complex<BFloat16>& operator-=(
+      const complex<BFloat16>& other) {
+    real_ = static_cast<float>(real_) - static_cast<float>(other.real_);
+    imag_ = static_cast<float>(imag_) - static_cast<float>(other.imag_);
+    return *this;
+  }
+
+  C10_HOST_DEVICE complex<BFloat16>& operator*=(
+      const complex<BFloat16>& other) {
+    auto a = static_cast<float>(real_);
+    auto b = static_cast<float>(imag_);
+    auto c = static_cast<float>(other.real());
+    auto d = static_cast<float>(other.imag());
+    real_ = a * c - b * d;
+    imag_ = a * d + b * c;
+    return *this;
+  }
+};
+
 } // namespace c10
 
 HIDDEN_NAMESPACE_BEGIN(torch, headeronly)
diff --git a/tools/cmake/Utils.cmake b/tools/cmake/Utils.cmake
index 958b425c47c..7219bf30ab4 100644
--- a/tools/cmake/Utils.cmake
+++ b/tools/cmake/Utils.cmake
@@ -142,6 +142,10 @@ macro(find_package_torch)
     add_torch_to_cmake_prefix_path()
     find_package(Torch CONFIG REQUIRED)
   endif()
+  # PyTorch sets CXX_STANDARD on its imported target, which does not propagate
+  # to consumers. Advertise the requirement as an interface compile feature so
+  # only targets that link Torch are compiled as C++20.
+  target_compile_features(torch INTERFACE cxx_std_20)
 endmacro()
 
 # Modify ${targetName}'s INTERFACE_INCLUDE_DIRECTORIES by wrapping each entry in
diff --git a/torch_pin.py b/torch_pin.py
index ca593b1ef05..45f0b9a20f2 100644
--- a/torch_pin.py
+++ b/torch_pin.py
@@ -1,2 +1,2 @@
 TORCH_VERSION = "2.13.0"
-# NIGHTLY_VERSION = "dev20260318" Temporarily pinning to stable release candidate. Revert https://github.com/pytorch/executorch/pull/18287
+NIGHTLY_VERSION = "dev20260809"