diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 62220467b71..25352f42c64 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -83,12 +83,12 @@ repos:
 
       - id: stubgen-pyx-cuda-core
         name: Generate .pyi stubs for cuda_core
-        entry: python ./toolshed/run_stubgen_pyx.py
+        entry: python -X utf8 -m stubgen_pyx cuda_core/cuda --continue-on-error --include-private
         language: python
         files: ^cuda_core/cuda/.*\.(pyx|pxd)$
         pass_filenames: false
         additional_dependencies:
-          - stubgen-pyx==0.2.6
+          - stubgen-pyx==0.2.19
           - Cython==3.2.9
 
   # Link checking for authored documentation files
diff --git a/cuda_core/cuda/core/_context.pyi b/cuda_core/cuda/core/_context.pyi
index afbc130882e..4adccfcbcfb 100644
--- a/cuda_core/cuda/core/_context.pyi
+++ b/cuda_core/cuda/core/_context.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_context.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_context.pyx
 
 from collections.abc import Sequence
 from dataclasses import dataclass
@@ -10,6 +8,8 @@ from cuda.core._device_resources import (DeviceResources, SMResource,
                                          WorkqueueResource)
 from cuda.core._stream import Stream
 
+__all__ = ['Context', 'ContextOptions']
+DeviceResourcesType = Sequence[SMResource | WorkqueueResource]
 
 class Context:
     """CUDA context wrapper.
@@ -17,25 +17,15 @@ class Context:
     Context objects represent CUDA contexts and cannot be instantiated directly.
     Use Device or Stream APIs to obtain context objects.
     """
-
-    def close(self):
-        """Release this context wrapper's underlying CUDA handles."""
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @property
     def handle(self) -> cuda.bindings.driver.CUcontext | None:
         """Return the underlying CUcontext handle."""
-
     @property
-    def _handle(self) -> cuda.bindings.driver.CUcontext | None:
-        ...
-
+    def _handle(self) -> cuda.bindings.driver.CUcontext | None: ...
     @property
     def is_green(self) -> bool:
         """True if this context was created from device resources."""
-
     @property
     def resources(self) -> DeviceResources:
         """Query the hardware resources provisioned for this context.
@@ -46,8 +36,7 @@ class Context:
 
         Raises :class:`RuntimeError` if the context has been closed.
         """
-
-    def create_stream(self, options: object=None) -> Stream:
+    def create_stream(self, options: object | None=None) -> Stream:
         """Create a new stream bound to this green context.
 
         This method is only available on green contexts. For primary
@@ -63,15 +52,11 @@ class Context:
         :obj:`~_stream.Stream`
             Newly created stream object.
         """
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
-    def __repr__(self) -> str:
-        ...
+    def close(self):
+        """Release this context wrapper's underlying CUDA handles."""
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
+    def __repr__(self) -> str: ...
 
 @dataclass
 class ContextOptions:
@@ -83,5 +68,3 @@ class ContextOptions:
         Device resources used to create a green context.
     """
     resources: DeviceResourcesType
-__all__ = ['Context', 'ContextOptions']
-DeviceResourcesType = Sequence[SMResource | WorkqueueResource]
\ No newline at end of file
diff --git a/cuda_core/cuda/core/_device.pyi b/cuda_core/cuda/core/_device.pyi
index e83aef8a8d0..825d7dd8cbc 100644
--- a/cuda_core/cuda/core/_device.pyi
+++ b/cuda_core/cuda/core/_device.pyi
@@ -1,8 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_device.pyx
-
-from __future__ import annotations
-
-import threading
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_device.pyx
 
 import cuda.core.system
 from cuda.core._context import Context, ContextOptions
@@ -17,6 +13,7 @@ from cuda.core.texture import (MipmappedArray, MipmappedArrayOptions,
                                ResourceDescriptor, SurfaceObject,
                                TextureObject, TextureObjectOptions)
 
+__all__ = ['Device']
 
 class DeviceProperties:
     """
@@ -24,588 +21,441 @@ class DeviceProperties:
 
     Attributes are read-only and provide information about the device.
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @classmethod
-    def _init(cls, handle: int) -> DeviceProperties:
-        ...
-
+    def _init(cls, handle: int) -> DeviceProperties: ...
     @property
     def max_threads_per_block(self) -> int:
         """int: Maximum number of threads per block."""
-
     @property
     def max_block_dim_x(self) -> int:
         """int: Maximum block dimension X."""
-
     @property
     def max_block_dim_y(self) -> int:
         """int: Maximum block dimension Y."""
-
     @property
     def max_block_dim_z(self) -> int:
         """int: Maximum block dimension Z."""
-
     @property
     def max_grid_dim_x(self) -> int:
         """int: Maximum grid dimension X."""
-
     @property
     def max_grid_dim_y(self) -> int:
         """int: Maximum grid dimension Y."""
-
     @property
     def max_grid_dim_z(self) -> int:
         """int: Maximum grid dimension Z."""
-
     @property
     def max_shared_memory_per_block(self) -> int:
         """int: Maximum shared memory available per block in bytes."""
-
     @property
     def total_constant_memory(self) -> int:
         """int: Memory available on device for constant variables in a CUDA C kernel in bytes."""
-
     @property
     def warp_size(self) -> int:
         """int: Warp size in threads."""
-
     @property
     def max_pitch(self) -> int:
         """int: Maximum pitch in bytes allowed by memory copies."""
-
     @property
     def maximum_texture1d_width(self) -> int:
         """int: Maximum 1D texture width."""
-
     @property
     def maximum_texture1d_linear_width(self) -> int:
         """int: Maximum width for a 1D texture bound to linear memory."""
-
     @property
     def maximum_texture1d_mipmapped_width(self) -> int:
         """int: Maximum mipmapped 1D texture width."""
-
     @property
     def maximum_texture2d_width(self) -> int:
         """int: Maximum 2D texture width."""
-
     @property
     def maximum_texture2d_height(self) -> int:
         """int: Maximum 2D texture height."""
-
     @property
     def maximum_texture2d_linear_width(self) -> int:
         """int: Maximum width for a 2D texture bound to linear memory."""
-
     @property
     def maximum_texture2d_linear_height(self) -> int:
         """int: Maximum height for a 2D texture bound to linear memory."""
-
     @property
     def maximum_texture2d_linear_pitch(self) -> int:
         """int: Maximum pitch in bytes for a 2D texture bound to linear memory."""
-
     @property
     def maximum_texture2d_mipmapped_width(self) -> int:
         """int: Maximum mipmapped 2D texture width."""
-
     @property
     def maximum_texture2d_mipmapped_height(self) -> int:
         """int: Maximum mipmapped 2D texture height."""
-
     @property
     def maximum_texture3d_width(self) -> int:
         """int: Maximum 3D texture width."""
-
     @property
     def maximum_texture3d_height(self) -> int:
         """int: Maximum 3D texture height."""
-
     @property
     def maximum_texture3d_depth(self) -> int:
         """int: Maximum 3D texture depth."""
-
     @property
     def maximum_texture3d_width_alternate(self) -> int:
         """int: Alternate maximum 3D texture width, 0 if no alternate maximum 3D texture size is supported."""
-
     @property
     def maximum_texture3d_height_alternate(self) -> int:
         """int: Alternate maximum 3D texture height, 0 if no alternate maximum 3D texture size is supported."""
-
     @property
     def maximum_texture3d_depth_alternate(self) -> int:
         """int: Alternate maximum 3D texture depth, 0 if no alternate maximum 3D texture size is supported."""
-
     @property
     def maximum_texturecubemap_width(self) -> int:
         """int: Maximum cubemap texture width or height."""
-
     @property
     def maximum_texture1d_layered_width(self) -> int:
         """int: Maximum 1D layered texture width."""
-
     @property
     def maximum_texture1d_layered_layers(self) -> int:
         """int: Maximum layers in a 1D layered texture."""
-
     @property
     def maximum_texture2d_layered_width(self) -> int:
         """int: Maximum 2D layered texture width."""
-
     @property
     def maximum_texture2d_layered_height(self) -> int:
         """int: Maximum 2D layered texture height."""
-
     @property
     def maximum_texture2d_layered_layers(self) -> int:
         """int: Maximum layers in a 2D layered texture."""
-
     @property
     def maximum_texturecubemap_layered_width(self) -> int:
         """int: Maximum cubemap layered texture width or height."""
-
     @property
     def maximum_texturecubemap_layered_layers(self) -> int:
         """int: Maximum layers in a cubemap layered texture."""
-
     @property
     def maximum_surface1d_width(self) -> int:
         """int: Maximum 1D surface width."""
-
     @property
     def maximum_surface2d_width(self) -> int:
         """int: Maximum 2D surface width."""
-
     @property
     def maximum_surface2d_height(self) -> int:
         """int: Maximum 2D surface height."""
-
     @property
     def maximum_surface3d_width(self) -> int:
         """int: Maximum 3D surface width."""
-
     @property
     def maximum_surface3d_height(self) -> int:
         """int: Maximum 3D surface height."""
-
     @property
     def maximum_surface3d_depth(self) -> int:
         """int: Maximum 3D surface depth."""
-
     @property
     def maximum_surface1d_layered_width(self) -> int:
         """int: Maximum 1D layered surface width."""
-
     @property
     def maximum_surface1d_layered_layers(self) -> int:
         """int: Maximum layers in a 1D layered surface."""
-
     @property
     def maximum_surface2d_layered_width(self) -> int:
         """int: Maximum 2D layered surface width."""
-
     @property
     def maximum_surface2d_layered_height(self) -> int:
         """int: Maximum 2D layered surface height."""
-
     @property
     def maximum_surface2d_layered_layers(self) -> int:
         """int: Maximum layers in a 2D layered surface."""
-
     @property
     def maximum_surfacecubemap_width(self) -> int:
         """int: Maximum cubemap surface width."""
-
     @property
     def maximum_surfacecubemap_layered_width(self) -> int:
         """int: Maximum cubemap layered surface width."""
-
     @property
     def maximum_surfacecubemap_layered_layers(self) -> int:
         """int: Maximum layers in a cubemap layered surface."""
-
     @property
     def max_registers_per_block(self) -> int:
         """int: Maximum number of 32-bit registers available to a thread block."""
-
     @property
     def clock_rate(self) -> int:
         """int: Typical clock frequency in kilohertz."""
-
     @property
     def texture_alignment(self) -> int:
         """int: Alignment requirement for textures."""
-
     @property
     def texture_pitch_alignment(self) -> int:
         """int: Pitch alignment requirement for textures."""
-
     @property
     def gpu_overlap(self) -> bool:
         """bool: Device can possibly copy memory and execute a kernel concurrently. Deprecated. Use :attr:`~DeviceProperties.async_engine_count` instead."""
-
     @property
     def multiprocessor_count(self) -> int:
         """int: Number of multiprocessors on device."""
-
     @property
     def kernel_exec_timeout(self) -> bool:
         """bool: Specifies whether there is a run time limit on kernels."""
-
     @property
     def integrated(self) -> bool:
         """bool: Device is integrated with host memory."""
-
     @property
     def can_map_host_memory(self) -> bool:
         """bool: Device can map host memory into CUDA address space."""
-
     @property
     def compute_mode(self) -> int:
         """int: Compute mode (See CUcomputemode for details)."""
-
     @property
     def concurrent_kernels(self) -> bool:
         """bool: Device can possibly execute multiple kernels concurrently."""
-
     @property
     def ecc_enabled(self) -> bool:
         """bool: Device has ECC support enabled."""
-
     @property
     def pci_bus_id(self) -> int:
         """int: PCI bus ID of the device."""
-
     @property
     def pci_device_id(self) -> int:
         """int: PCI device ID of the device."""
-
     @property
     def pci_domain_id(self) -> int:
         """int: PCI domain ID of the device."""
-
     @property
     def tcc_driver(self) -> bool:
         """bool: Device is using TCC driver model."""
-
     @property
     def memory_clock_rate(self) -> int:
         """int: Peak memory clock frequency in kilohertz."""
-
     @property
     def global_memory_bus_width(self) -> int:
         """int: Global memory bus width in bits."""
-
     @property
     def l2_cache_size(self) -> int:
         """int: Size of L2 cache in bytes."""
-
     @property
     def max_threads_per_multiprocessor(self) -> int:
         """int: Maximum resident threads per multiprocessor."""
-
     @property
     def unified_addressing(self) -> bool:
         """bool: Device shares a unified address space with the host."""
-
     @property
     def compute_capability_major(self) -> int:
         """int: Major compute capability version number."""
-
     @property
     def compute_capability_minor(self) -> int:
         """int: Minor compute capability version number."""
-
     @property
     def global_l1_cache_supported(self) -> bool:
         """bool: Device supports caching globals in L1."""
-
     @property
     def local_l1_cache_supported(self) -> bool:
         """bool: Device supports caching locals in L1."""
-
     @property
     def max_shared_memory_per_multiprocessor(self) -> int:
         """int: Maximum shared memory available per multiprocessor in bytes."""
-
     @property
     def max_registers_per_multiprocessor(self) -> int:
         """int: Maximum number of 32-bit registers available per multiprocessor."""
-
     @property
     def managed_memory(self) -> bool:
         """bool: Device can allocate managed memory on this system."""
-
     @property
     def multi_gpu_board(self) -> bool:
         """bool: Device is on a multi-GPU board."""
-
     @property
     def multi_gpu_board_group_id(self) -> int:
         """int: Unique id for a group of devices on the same multi-GPU board."""
-
     @property
     def host_native_atomic_supported(self) -> bool:
         """bool: Link between the device and the host supports all native atomic operations."""
-
     @property
     def single_to_double_precision_perf_ratio(self) -> int:
         """int: Ratio of single precision performance (in floating-point operations per second) to double precision performance."""
-
     @property
     def pageable_memory_access(self) -> bool:
         """bool: Device supports coherently accessing pageable memory without calling cudaHostRegister on it."""
-
     @property
     def concurrent_managed_access(self) -> bool:
         """bool: Device can coherently access managed memory concurrently with the CPU."""
-
     @property
     def compute_preemption_supported(self) -> bool:
         """bool: Device supports compute preemption."""
-
     @property
     def can_use_host_pointer_for_registered_mem(self) -> bool:
         """bool: Device can access host registered memory at the same virtual address as the CPU."""
-
     @property
     def cooperative_launch(self) -> bool:
         """bool: Device supports launching cooperative kernels via cuLaunchCooperativeKernel."""
-
     @property
     def max_shared_memory_per_block_optin(self) -> int:
         """int: Maximum optin shared memory per block."""
-
     @property
     def pageable_memory_access_uses_host_page_tables(self) -> bool:
         """bool: Device accesses pageable memory via the host's page tables."""
-
     @property
     def direct_managed_mem_access_from_host(self) -> bool:
         """bool: The host can directly access managed memory on the device without migration."""
-
     @property
     def virtual_memory_management_supported(self) -> bool:
         """bool: Device supports virtual memory management APIs like cuMemAddressReserve, cuMemCreate, cuMemMap and related APIs."""
-
     @property
     def handle_type_posix_file_descriptor_supported(self) -> bool:
         """bool: Device supports exporting memory to a posix file descriptor with cuMemExportToShareableHandle, if requested via cuMemCreate."""
-
     @property
     def handle_type_win32_handle_supported(self) -> bool:
         """bool: Device supports exporting memory to a Win32 NT handle with cuMemExportToShareableHandle, if requested via cuMemCreate."""
-
     @property
     def handle_type_win32_kmt_handle_supported(self) -> bool:
         """bool: Device supports exporting memory to a Win32 KMT handle with cuMemExportToShareableHandle, if requested via cuMemCreate."""
-
     @property
     def max_blocks_per_multiprocessor(self) -> int:
         """int: Maximum number of blocks per multiprocessor."""
-
     @property
     def generic_compression_supported(self) -> bool:
         """bool: Device supports compression of memory."""
-
     @property
     def max_persisting_l2_cache_size(self) -> int:
         """int: Maximum L2 persisting lines capacity setting in bytes."""
-
     @property
     def max_access_policy_window_size(self) -> int:
         """int: Maximum value of CUaccessPolicyWindow.num_bytes."""
-
     @property
     def gpu_direct_rdma_with_cuda_vmm_supported(self) -> bool:
         """bool: Device supports specifying the GPUDirect RDMA flag with cuMemCreate."""
-
     @property
     def reserved_shared_memory_per_block(self) -> int:
         """int: Shared memory reserved by CUDA driver per block in bytes."""
-
     @property
     def sparse_cuda_array_supported(self) -> bool:
         """bool: Device supports sparse CUDA arrays and sparse CUDA mipmapped arrays."""
-
     @property
     def read_only_host_register_supported(self) -> bool:
         """bool: True if device supports using the cuMemHostRegister flag CU_MEMHOSTREGISTER_READ_ONLY to register memory that must be mapped as read-only to the GPU, False if not."""
-
     @property
     def memory_pools_supported(self) -> bool:
         """bool: Device supports using the cuMemAllocAsync and cuMemPool family of APIs."""
-
     @property
     def gpu_direct_rdma_supported(self) -> bool:
         """bool: Device supports GPUDirect RDMA APIs, like nvidia_p2p_get_pages (see https://docs.nvidia.com/cuda/gpudirect-rdma for more information)."""
-
     @property
     def gpu_direct_rdma_flush_writes_options(self) -> int:
         """int: The returned attribute shall be interpreted as a bitmask, where the individual bits are described by the CUflushGPUDirectRDMAWritesOptions enum."""
-
     @property
     def gpu_direct_rdma_writes_ordering(self) -> int:
         """int: GPUDirect RDMA writes to the device do not need to be flushed for consumers within the scope indicated by the returned attribute. See CUGPUDirectRDMAWritesOrdering for the numerical values returned here."""
-
     @property
     def mempool_supported_handle_types(self) -> int:
         """int: Handle types supported with mempool based IPC."""
-
     @property
     def deferred_mapping_cuda_array_supported(self) -> bool:
         """bool: Device supports deferred mapping CUDA arrays and CUDA mipmapped arrays."""
-
     @property
     def numa_config(self) -> int:
         """int: NUMA configuration of a device: value is of type CUdeviceNumaConfig enum."""
-
     @property
     def numa_id(self) -> int:
         """int: NUMA node ID of the GPU memory."""
-
     @property
     def multicast_supported(self) -> bool:
         """bool: Device supports switch multicast and reduction operations."""
-
     @property
     def surface_alignment(self) -> int:
         """int: Surface alignment requirement in bytes."""
-
     @property
     def async_engine_count(self) -> int:
         """int: Number of asynchronous engines."""
-
     @property
     def can_tex2d_gather(self) -> bool:
         """bool: True if device supports 2D texture gather operations, False if not."""
-
     @property
     def maximum_texture2d_gather_width(self) -> int:
         """int: Maximum 2D texture gather width."""
-
     @property
     def maximum_texture2d_gather_height(self) -> int:
         """int: Maximum 2D texture gather height."""
-
     @property
     def stream_priorities_supported(self) -> bool:
         """bool: True if device supports stream priorities, False if not."""
-
     @property
     def can_flush_remote_writes(self) -> bool:
         """bool: The CU_STREAM_WAIT_VALUE_FLUSH flag and the CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES MemOp are supported on the device. See Stream Memory Operations for additional details."""
-
     @property
     def host_register_supported(self) -> bool:
         """bool: Device supports host memory registration via cudaHostRegister."""
-
     @property
     def timeline_semaphore_interop_supported(self) -> bool:
         """bool: External timeline semaphore interop is supported on the device."""
-
     @property
     def cluster_launch(self) -> bool:
         """bool: Indicates device supports cluster launch."""
-
     @property
     def can_use_64_bit_stream_mem_ops(self) -> bool:
         """bool: 64-bit operations are supported in cuStreamBatchMemOp and related MemOp APIs."""
-
     @property
     def can_use_stream_wait_value_nor(self) -> bool:
         """bool: CU_STREAM_WAIT_VALUE_NOR is supported by MemOp APIs."""
-
     @property
     def dma_buf_supported(self) -> bool:
         """bool: Device supports buffer sharing with dma_buf mechanism."""
-
     @property
     def ipc_event_supported(self) -> bool:
         """bool: Device supports IPC Events."""
-
     @property
     def mem_sync_domain_count(self) -> int:
         """int: Number of memory domains the device supports."""
-
     @property
     def tensor_map_access_supported(self) -> bool:
         """bool: Device supports accessing memory using Tensor Map."""
-
     @property
     def handle_type_fabric_supported(self) -> bool:
         """bool: Device supports exporting memory to a fabric handle with cuMemExportToShareableHandle() or requested with cuMemCreate()."""
-
     @property
     def unified_function_pointers(self) -> bool:
         """bool: Device supports unified function pointers."""
-
     @property
     def mps_enabled(self) -> bool:
         """bool: Indicates if contexts created on this device will be shared via MPS."""
-
     @property
     def host_numa_id(self) -> int:
         """int: NUMA ID of the host node closest to the device. Returns -1 when system does not support NUMA."""
-
     @property
     def d3d12_cig_supported(self) -> bool:
         """bool: Device supports CIG with D3D12."""
-
     @property
     def mem_decompress_algorithm_mask(self) -> int:
         """int: The returned value shall be interpreted as a bitmask, where the individual bits are described by the CUmemDecompressAlgorithm enum."""
-
     @property
     def mem_decompress_maximum_length(self) -> int:
         """int: The returned value is the maximum length in bytes of a single decompress operation that is allowed."""
-
     @property
     def vulkan_cig_supported(self) -> bool:
         """bool: Device supports CIG with Vulkan."""
-
     @property
     def gpu_pci_device_id(self) -> int:
         """int: The combined 16-bit PCI device ID and 16-bit PCI vendor ID.
 
         Returns 0 if the driver does not support this query.
         """
-
     @property
     def gpu_pci_subsystem_id(self) -> int:
         """int: The combined 16-bit PCI subsystem ID and 16-bit PCI subsystem vendor ID.
 
         Returns 0 if the driver does not support this query.
         """
-
     @property
     def host_numa_virtual_memory_management_supported(self) -> bool:
         """bool: Device supports HOST_NUMA location with the virtual memory management APIs like cuMemCreate, cuMemMap and related APIs."""
-
     @property
     def host_numa_memory_pools_supported(self) -> bool:
         """bool: Device supports HOST_NUMA location with the cuMemAllocAsync and cuMemPool family of APIs."""
-
     @property
     def host_numa_multinode_ipc_supported(self) -> bool:
         """bool: Device supports HOST_NUMA location IPC between nodes in a multi-node system."""
-
     @property
     def host_memory_pools_supported(self) -> bool:
         """bool: Device supports HOST location with the cuMemAllocAsync and cuMemPool family of APIs."""
-
     @property
     def host_virtual_memory_management_supported(self) -> bool:
         """bool: Device supports HOST location with the virtual memory management APIs like cuMemCreate, cuMemMap and related APIs."""
-
     @property
     def host_alloc_dma_buf_supported(self) -> bool:
         """bool: Device supports page-locked host memory buffer sharing with dma_buf mechanism."""
-
     @property
     def only_partial_host_native_atomic_supported(self) -> bool:
         """bool: Link between the device and the host supports only some native atomic operations."""
@@ -638,12 +488,8 @@ class Device:
     """
     __slots__ = ('_device_id', '_memory_resource', '_has_inited', '_properties', '_resources', '_uuid', '_context', '__weakref__')
 
-    def __new__(cls, device_id: Device | int | None=None) -> Device:
-        ...
-
-    def _check_context_initialized(self) -> None:
-        ...
-
+    def __new__(cls, device_id: Device | int | None=None) -> Device: ...
+    def _check_context_initialized(self) -> None: ...
     @classmethod
     def get_all_devices(cls) -> tuple[Device, ...]:
         """
@@ -654,8 +500,7 @@ class Device:
         tuple of Device
             A tuple containing instances of available devices.
         """
-
-    def to_system_device(self) -> 'cuda.core.system.Device':
+    def to_system_device(self) -> cuda.core.system.Device:
         """
         Get the corresponding :class:`cuda.core.system.Device` (which is used
         for NVIDIA Management Library (NVML) access) for this
@@ -668,15 +513,12 @@ class Device:
         cuda.core.system.Device
             The corresponding system-level device instance used for NVML access.
         """
-
     @property
     def device_id(self) -> int:
         """Return device ordinal."""
-
     @property
     def pci_bus_id(self) -> str:
         """Return a PCI Bus Id string for this device."""
-
     def can_access_peer(self, peer: Device | int) -> bool:
         """Check if this device can access memory from the specified peer device.
 
@@ -688,7 +530,6 @@ class Device:
         peer : Device | int
             The peer device to check accessibility to. Can be a :obj:`~_device.Device` object or device ID.
         """
-
     @property
     def uuid(self) -> str:
         """Return a UUID for the device.
@@ -705,27 +546,21 @@ class Device:
         The UUID is cached after first access to avoid repeated CUDA API calls.
 
         """
-
     @property
     def name(self) -> str:
         """Return the device name."""
-
     @property
     def properties(self) -> DeviceProperties:
         """Return a :obj:`~_device.DeviceProperties` class with information about the device."""
-
     @property
     def resources(self) -> DeviceResources:
         """Return the hardware resource query namespace for this device."""
-
     @property
     def compute_capability(self) -> ComputeCapability:
         """Return a named tuple with 2 fields: major and minor."""
-
     @property
     def arch(self) -> str:
         """Return compute capability as a string (e.g., '75' for CC 7.5)."""
-
     @property
     def context(self) -> Context:
         """Return the :obj:`~_context.Context` associated with this device.
@@ -735,15 +570,11 @@ class Device:
         Device must be initialized.
 
         """
-
     @property
     def memory_resource(self) -> MemoryResource:
         """Return :obj:`~_memory.MemoryResource` associated with this device."""
-
     @memory_resource.setter
-    def memory_resource(self, mr: MemoryResource) -> None:
-        ...
-
+    def memory_resource(self, mr: MemoryResource) -> None: ...
     @property
     def default_stream(self) -> Stream:
         """Return default CUDA :obj:`~_stream.Stream` associated with this device.
@@ -755,22 +586,12 @@ class Device:
         the legacy stream.
 
         """
-
     def __int__(self) -> int:
         """Return device_id."""
-
-    def __repr__(self) -> str:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-
+    def __repr__(self) -> str: ...
+    def __hash__(self) -> int: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __reduce__(self) -> tuple[object, ...]: ...
     def set_current(self, ctx: Context | None=None) -> Context | None:
         """Set device to be used for GPU executions.
 
@@ -801,7 +622,6 @@ class Device:
         >>> # ... do work on device 0 ...
 
         """
-
     def create_context(self, options: ContextOptions | None=None) -> Context:
         """Create a new :obj:`~_context.Context` object.
 
@@ -820,8 +640,7 @@ class Device:
             Newly created context object.
 
         """
-
-    def create_stream(self, obj: IsStreamType | None=None, options: object=None) -> Stream:
+    def create_stream(self, obj: IsStreamType | None=None, options: object | None=None) -> Stream:
         """Create a :obj:`~_stream.Stream` object.
 
         New stream objects can be created in two different ways:
@@ -849,7 +668,6 @@ class Device:
             Newly created stream object.
 
         """
-
     def create_event(self, options: EventOptions | None=None) -> Event:
         """Create an :obj:`~_event.Event` object without recording it to a :obj:`~_stream.Stream`.
 
@@ -868,7 +686,6 @@ class Device:
             Newly created event object.
 
         """
-
     def allocate(self, size: int, *, stream: Stream | GraphBuilder) -> Buffer:
         """Allocate device memory from a specified stream.
 
@@ -894,7 +711,6 @@ class Device:
             Newly created buffer object.
 
         """
-
     def sync(self) -> None:
         """Synchronize the device.
 
@@ -903,7 +719,6 @@ class Device:
         Device must be initialized.
 
         """
-
     def create_graph_builder(self) -> GraphBuilder:
         """Create a new :obj:`~graph.GraphBuilder` object.
 
@@ -913,7 +728,6 @@ class Device:
             Newly created graph builder object.
 
         """
-
     def create_opaque_array(self, options: OpaqueArrayOptions) -> OpaqueArray:
         """Create an :obj:`~cuda.core.texture.OpaqueArray` on the current device.
 
@@ -938,7 +752,6 @@ class Device:
 
         .. versionadded:: 1.1.0
         """
-
     def create_mipmapped_array(self, options: MipmappedArrayOptions) -> MipmappedArray:
         """Create a :obj:`~cuda.core.texture.MipmappedArray` on the current device.
 
@@ -963,7 +776,6 @@ class Device:
 
         .. versionadded:: 1.1.0
         """
-
     def create_texture_object(self, *, resource: ResourceDescriptor, options: TextureObjectOptions | None=None) -> TextureObject:
         """Create a :obj:`~cuda.core.texture.TextureObject` on the current device.
 
@@ -993,7 +805,6 @@ class Device:
 
         .. versionadded:: 1.1.0
         """
-
     def create_surface_object(self, *, resource: ResourceDescriptor) -> SurfaceObject:
         """Create a :obj:`~cuda.core.texture.SurfaceObject` on the current device.
 
@@ -1022,6 +833,3 @@ class Device:
 
         .. versionadded:: 1.1.0
         """
-_tls = threading.local()
-_lock = threading.Lock()
-__all__ = ['Device']
\ No newline at end of file
diff --git a/cuda_core/cuda/core/_device_resources.pyi b/cuda_core/cuda/core/_device_resources.pyi
index 7514f5a2f43..a6837e837e9 100644
--- a/cuda_core/cuda/core/_device_resources.pyi
+++ b/cuda_core/cuda/core/_device_resources.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_device_resources.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_device_resources.pyx
 
 from collections.abc import Sequence as SequenceABC
 from dataclasses import dataclass
@@ -8,6 +6,7 @@ from dataclasses import dataclass
 from cuda.core._device import Device
 from cuda.core.typing import WorkqueueSharingScopeType
 
+__all__ = ['DeviceResources', 'SMResource', 'SMResourceOptions', 'WorkqueueResource', 'WorkqueueResourceOptions']
 
 @dataclass
 class SMResourceOptions:
@@ -61,8 +60,7 @@ class WorkqueueResourceOptions:
     sharing_scope: WorkqueueSharingScopeType | str | None = None
     concurrency_limit: int | None = None
 
-    def __post_init__(self):
-        ...
+    def __post_init__(self): ...
 
 class SMResource:
     """Represent an SM (streaming multiprocessor) resource partition.
@@ -70,30 +68,22 @@ class SMResource:
     Instances are returned by :obj:`DeviceResources.sm` or
     :meth:`SMResource.split` and cannot be instantiated directly.
     """
-
-    def __init__(self, *args, **kwargs):
-        ...
-
+    def __init__(self, *args, **kwargs): ...
     @property
     def handle(self) -> int:
         """Return the address of the underlying ``CUdevResource`` struct."""
-
     @property
     def sm_count(self) -> int:
         """Total SMs available in this resource."""
-
     @property
     def min_partition_size(self) -> int:
         """Minimum SM count required to create a partition."""
-
     @property
     def coscheduled_alignment(self) -> int:
         """Number of SMs guaranteed to be co-scheduled."""
-
     @property
     def flags(self) -> int:
         """Raw flags from the underlying SM resource."""
-
     def split(self, options: SMResourceOptions, *, dry_run: bool=False) -> tuple[list[SMResource], SMResource]:
         """Split this SM resource into groups and a remainder.
 
@@ -120,14 +110,10 @@ class WorkqueueResource:
     Instances are returned by :obj:`DeviceResources.workqueue` and
     cannot be instantiated directly.
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @property
     def handle(self) -> int:
         """Return the address of the underlying config ``CUdevResource`` struct."""
-
     @property
     def sharing_scope(self) -> WorkqueueSharingScopeType:
         """Current sharing scope of this workqueue resource.
@@ -138,7 +124,6 @@ class WorkqueueResource:
         :meth:`configure` with
         :attr:`WorkqueueResourceOptions.sharing_scope`.
         """
-
     @property
     def concurrency_limit(self) -> int:
         """Current expected maximum concurrent stream-ordered workloads.
@@ -149,11 +134,9 @@ class WorkqueueResource:
         via :meth:`configure` with
         :attr:`WorkqueueResourceOptions.concurrency_limit`.
         """
-
     @property
     def device(self) -> Device:
         """The :class:`~cuda.core.Device` this workqueue resource is available on."""
-
     def configure(self, options: WorkqueueResourceOptions) -> None:
         """Configure the workqueue resource in place.
 
@@ -173,15 +156,10 @@ class DeviceResources:
 
     This class cannot be instantiated directly.
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @property
     def sm(self) -> SMResource:
         """Return the :obj:`SMResource` for this device or context."""
-
     @property
     def workqueue(self) -> WorkqueueResource:
         """Return the :obj:`WorkqueueResource` for this device or context."""
-__all__ = ['DeviceResources', 'SMResource', 'SMResourceOptions', 'WorkqueueResource', 'WorkqueueResourceOptions']
\ No newline at end of file
diff --git a/cuda_core/cuda/core/_dlpack.pyi b/cuda_core/cuda/core/_dlpack.pyi
index 575d9ced8f5..37d65d9f0e4 100644
--- a/cuda_core/cuda/core/_dlpack.pyi
+++ b/cuda_core/cuda/core/_dlpack.pyi
@@ -1,11 +1,10 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_dlpack.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_dlpack.pyx
 
 from enum import IntEnum
+from typing import Any, Callable, TypeAlias, TypedDict
 
-_DLDeviceType = int
-DLDataTypeCode = int
+_DLDeviceType: TypeAlias = int
+DLDataTypeCode: TypeAlias = int
 
 class DLDeviceType(IntEnum):
     kDLCPU = 1
@@ -13,12 +12,56 @@ class DLDeviceType(IntEnum):
     kDLCUDAHost = 3
     kDLCUDAManaged = 13
 
-def make_py_capsule(buf: object, versioned: bool) -> object:
-    ...
+class DLDevice(TypedDict):
+    device_type: _DLDeviceType
+    device_id: int
+
+class DLDataType(TypedDict):
+    code: int
+    bits: int
+    lanes: int
+
+class DLTensor(TypedDict):
+    data: None
+    device: DLDevice
+    ndim: int
+    dtype: DLDataType
+    shape: int
+    strides: int
+    byte_offset: int
+
+class DLManagedTensor(TypedDict):
+    dl_tensor: DLTensor
+    manager_ctx: None
+    deleter: Callable[..., Any]
+
+class DLPackVersion(TypedDict):
+    major: int
+    minor: int
+
+class DLManagedTensorVersioned(TypedDict):
+    version: DLPackVersion
+    manager_ctx: None
+    deleter: Callable[..., Any]
+    flags: int
+    dl_tensor: DLTensor
+
+class DLPackExchangeAPIHeader(TypedDict):
+    version: DLPackVersion
+    prev_api: DLPackExchangeAPIHeader
+
+class DLPackExchangeAPI(TypedDict):
+    header: DLPackExchangeAPIHeader
+    managed_tensor_allocator: ...
+    managed_tensor_from_py_object_no_sync: ...
+    managed_tensor_to_py_object_no_sync: ...
+    dltensor_from_py_object_no_sync: ...
+    current_work_stream: ...
 
 def classify_dl_device(buf: object) -> tuple[int, int]:
     """Classify a buffer into a DLPack (device_type, device_id) pair.
 
     ``buf`` must expose ``is_device_accessible``, ``is_host_accessible``,
     ``is_managed``, and ``device_id`` attributes.
-    """
\ No newline at end of file
+    """
+def make_py_capsule(buf: object, versioned: bool) -> object: ...
diff --git a/cuda_core/cuda/core/_event.pyi b/cuda_core/cuda/core/_event.pyi
index 9391735b6ab..3fbb1b0b1b0 100644
--- a/cuda_core/cuda/core/_event.pyi
+++ b/cuda_core/cuda/core/_event.pyi
@@ -1,14 +1,13 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_event.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_event.pyx
 
 from dataclasses import dataclass
+from typing import Any
 
 import cuda.bindings.driver
-import cython
 from cuda.core._context import Context
 from cuda.core._device import Device
 
+__all__ = ['Event', 'EventOptions']
 
 @dataclass
 class EventOptions:
@@ -61,39 +60,22 @@ class Event:
     and they should instead be created through a :obj:`~_stream.Stream` object.
 
     """
-
+    def __init__(self, *args, **kwargs) -> None: ...
     def close(self):
         """Destroy the event.
 
         Releases the event handle. The underlying CUDA event is destroyed
         when the last reference is released.
         """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
-    def __isub__(self, other: object):
-        ...
-
-    def __rsub__(self, other: object):
-        ...
-
-    def __sub__(self, other: Event) -> float:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-
+    def __isub__(self, other: object): ...
+    def __rsub__(self, other: object): ...
+    def __sub__(self, other: Event) -> float: ...
+    def __hash__(self) -> int: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __repr__(self) -> str: ...
     @property
     def ipc_descriptor(self) -> IPCEventDescriptor:
         """Descriptor for sharing this event with other processes."""
-
     @classmethod
     def from_ipc_descriptor(cls, ipc_descriptor: IPCEventDescriptor) -> Event:
         """Import an event that was exported from another process.
@@ -110,21 +92,17 @@ class Event:
             A new event backed by the imported IPC handle.
 
         """
-
     @property
     def is_ipc_enabled(self) -> bool:
         """Return True if the event can be shared across process boundaries, otherwise False."""
-
     @property
     def is_timing_enabled(self) -> bool:
         """Return True if the event records timing data, otherwise False."""
-
     @property
     def is_blocking_sync(self) -> bool:
         """Return True if the event uses blocking synchronization (the CPU
         thread blocks on :meth:`sync` instead of busy-waiting), otherwise False.
         """
-
     def sync(self) -> None:
         """Synchronize until the event completes.
 
@@ -134,11 +112,9 @@ class Event:
         thread busy-waits until the event has completed.
 
         """
-
     @property
     def is_done(self) -> bool:
         """Return True if all captured works have been completed, otherwise False."""
-
     @property
     def handle(self) -> cuda.bindings.driver.CUevent:
         """Return the underlying CUevent object.
@@ -148,7 +124,6 @@ class Event:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(Event.handle)``.
         """
-
     @property
     def device(self) -> Device:
         """Return the :obj:`~_device.Device` singleton associated with this event.
@@ -160,27 +135,16 @@ class Event:
         context is set current after a event is created.
 
         """
-
     @property
     def context(self) -> Context:
         """Return the :obj:`~_context.Context` associated with this event."""
 
 class IPCEventDescriptor:
     """Serializable object describing an event that can be shared between processes."""
-
-    def __init__(self, *arg, **kwargs) -> None:
-        ...
-
+    def __init__(self, *arg, **kwargs) -> None: ...
     @staticmethod
-    def _init(reserved: bytes, is_blocking_sync: cython.bint) -> IPCEventDescriptor:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-__all__ = ['Event', 'EventOptions']
+    def _init(reserved: bytes, is_blocking_sync: Any) -> IPCEventDescriptor: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __reduce__(self) -> tuple[object, ...]: ...
 
-def _reduce_event(event: Event) -> tuple[object, ...]:
-    ...
\ No newline at end of file
+def _reduce_event(event: Event) -> tuple[object, ...]: ...
diff --git a/cuda_core/cuda/core/_graphics.pyi b/cuda_core/cuda/core/_graphics.pyi
index b7022e5a18a..3ca77a137b6 100644
--- a/cuda_core/cuda/core/_graphics.pyi
+++ b/cuda_core/cuda/core/_graphics.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_graphics.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_graphics.pyx
 
 from typing import Sequence
 
@@ -8,6 +6,8 @@ from cuda.bindings import cydriver
 from cuda.core._memory._buffer import Buffer
 from cuda.core._stream import Stream
 
+__all__ = ['GraphicsResource']
+_REGISTER_FLAGS = {'none': cydriver.CU_GRAPHICS_REGISTER_FLAGS_NONE, 'read_only': cydriver.CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY, 'write_discard': cydriver.CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD, 'surface_load_store': cydriver.CU_GRAPHICS_REGISTER_FLAGS_SURFACE_LDST, 'texture_gather': cydriver.CU_GRAPHICS_REGISTER_FLAGS_TEXTURE_GATHER}
 
 class GraphicsResource:
     """RAII wrapper for a CUDA graphics resource (``CUgraphicsResource``).
@@ -48,23 +48,7 @@ class GraphicsResource:
                 # ... launch kernels using buf.handle, buf.size ...
                 pass
     """
-
-    def close(self, stream: object=None):
-        """Unregister this graphics resource from CUDA.
-
-        If the resource is currently mapped, it is unmapped first. After
-        closing, the resource cannot be used again.
-
-        Parameters
-        ----------
-        stream : :class:`~cuda.core.Stream`, optional
-            Optional override for the stream used to close the currently
-            mapped buffer, if one exists.
-        """
-
-    def __init__(self) -> None:
-        ...
-
+    def __init__(self) -> None: ...
     @classmethod
     def from_gl_buffer(cls, gl_buffer: int, *, flags: str | tuple[str, ...] | list[str] | None=None, stream: Stream | None=None) -> GraphicsResource:
         """Register an OpenGL buffer object for CUDA access.
@@ -111,7 +95,6 @@ class GraphicsResource:
         ValueError
             If an unknown flag string is provided.
         """
-
     @classmethod
     def from_gl_image(cls, image: int, target: int, *, flags: str | tuple[str, ...] | list[str] | None=None) -> GraphicsResource:
         """Register an OpenGL texture or renderbuffer for CUDA access.
@@ -142,10 +125,7 @@ class GraphicsResource:
         ValueError
             If an unknown flag string is provided.
         """
-
-    def _get_mapped_buffer(self) -> object:
-        ...
-
+    def _get_mapped_buffer(self) -> object: ...
     def map(self, *, stream: Stream) -> Buffer:
         """Map this graphics resource for CUDA access.
 
@@ -178,7 +158,6 @@ class GraphicsResource:
         CUDAError
             If the mapping fails.
         """
-
     def unmap(self, *, stream: Stream | None=None) -> None:
         """Unmap this graphics resource, releasing it back to the graphics API.
 
@@ -198,29 +177,29 @@ class GraphicsResource:
         CUDAError
             If the unmapping fails.
         """
+    def __enter__(self) -> object: ...
+    def __exit__(self, exc_type: type | None, exc_val: BaseException | None, exc_tb: object) -> bool: ...
+    def close(self, stream: object | None=None):
+        """Unregister this graphics resource from CUDA.
 
-    def __enter__(self) -> object:
-        ...
-
-    def __exit__(self, exc_type: type | None, exc_val: BaseException | None, exc_tb: object) -> bool:
-        ...
+        If the resource is currently mapped, it is unmapped first. After
+        closing, the resource cannot be used again.
 
+        Parameters
+        ----------
+        stream : :class:`~cuda.core.Stream`, optional
+            Optional override for the stream used to close the currently
+            mapped buffer, if one exists.
+        """
     @property
     def is_mapped(self) -> bool:
         """Whether the resource is currently mapped for CUDA access."""
-
     @property
     def handle(self) -> int:
         """The raw ``CUgraphicsResource`` handle as a Python int."""
-
     @property
     def resource_handle(self) -> int:
         """Alias for :attr:`handle`."""
+    def __repr__(self) -> str: ...
 
-    def __repr__(self) -> str:
-        ...
-__all__ = ['GraphicsResource']
-_REGISTER_FLAGS = {'none': cydriver.CU_GRAPHICS_REGISTER_FLAGS_NONE, 'read_only': cydriver.CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY, 'write_discard': cydriver.CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD, 'surface_load_store': cydriver.CU_GRAPHICS_REGISTER_FLAGS_SURFACE_LDST, 'texture_gather': cydriver.CU_GRAPHICS_REGISTER_FLAGS_TEXTURE_GATHER}
-
-def _parse_register_flags(flags: str | Sequence[str] | None) -> int:
-    ...
\ No newline at end of file
+def _parse_register_flags(flags: str | Sequence[str] | None) -> int: ...
diff --git a/cuda_core/cuda/core/_kernel_arg_handler.pyi b/cuda_core/cuda/core/_kernel_arg_handler.pyi
index 0ebd2c0d0b6..918548ef40c 100644
--- a/cuda_core/cuda/core/_kernel_arg_handler.pyi
+++ b/cuda_core/cuda/core/_kernel_arg_handler.pyi
@@ -1,18 +1,15 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_kernel_arg_handler.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_kernel_arg_handler.pyx
 
-from __future__ import annotations
+from typing import Any, Sequence, TypeAlias, TypedDict
 
-from typing import Any, Sequence
-
-from libcpp.complex import complex as cpp_complex
+cpp_single_complex: TypeAlias = Any
+cpp_double_complex: TypeAlias = Any
 
+class __half_raw(TypedDict):
+    x: int
 
 class ParamHolder:
+    ptr: int
 
-    def __init__(self, kernel_args: Sequence[Any]) -> None:
-        ...
-
-    def __dealloc__(self) -> None:
-        ...
-cpp_single_complex = cpp_complex.complex
-cpp_double_complex = cpp_complex.complex
\ No newline at end of file
+    def __init__(self, kernel_args: Sequence[Any]) -> None: ...
+    def __dealloc__(self) -> None: ...
diff --git a/cuda_core/cuda/core/_launch_config.pyi b/cuda_core/cuda/core/_launch_config.pyi
index 47187fb03d6..a731f2999ff 100644
--- a/cuda_core/cuda/core/_launch_config.pyi
+++ b/cuda_core/cuda/core/_launch_config.pyi
@@ -1,9 +1,9 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_launch_config.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_launch_config.pyx
 
 from typing import Any
 
+_LAUNCH_CONFIG_ATTRS = ('grid', 'cluster', 'block', 'shmem_size', 'is_cooperative', 'programmatic_stream_serialization')
+__all__ = ['LaunchConfig']
 
 class LaunchConfig:
     """Customizable launch options.
@@ -40,6 +40,12 @@ class LaunchConfig:
         the kernel may overlap with a previous kernel in the same stream that
         signals completion via programmatic means.
     """
+    grid: tuple[Any, ...]
+    cluster: tuple[Any, ...]
+    block: tuple[Any, ...]
+    shmem_size: int
+    is_cooperative: bool
+    programmatic_stream_serialization: bool
 
     def __init__(self, grid: int | tuple[int, ...] | None=None, cluster: int | tuple[int, ...] | None=None, block: int | tuple[int, ...] | None=None, shmem_size: int | None=None, is_cooperative: bool=False, programmatic_stream_serialization: bool=False) -> None:
         """Initialize LaunchConfig with validation.
@@ -59,20 +65,11 @@ class LaunchConfig:
         programmatic_stream_serialization : bool, optional
             Whether to allow programmatic stream serialization / PDL (default: False)
         """
-
-    def _identity(self) -> tuple[Any, ...]:
-        ...
-
+    def _identity(self) -> tuple[Any, ...]: ...
     def __repr__(self) -> str:
         """Return string representation of LaunchConfig."""
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-_LAUNCH_CONFIG_ATTRS = ('grid', 'cluster', 'block', 'shmem_size', 'is_cooperative', 'programmatic_stream_serialization')
-__all__ = ['LaunchConfig']
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
 
 def _to_native_launch_config(config: LaunchConfig) -> object:
     """Convert LaunchConfig to native driver CUlaunchConfig.
@@ -86,4 +83,4 @@ def _to_native_launch_config(config: LaunchConfig) -> object:
     -------
     driver.CUlaunchConfig
         Native CUDA driver launch configuration
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/_launcher.pyi b/cuda_core/cuda/core/_launcher.pyi
index 27ed7e86da7..72ba7c6c623 100644
--- a/cuda_core/cuda/core/_launcher.pyi
+++ b/cuda_core/cuda/core/_launcher.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_launcher.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_launcher.pyx
 
 from cuda.core._launch_config import LaunchConfig
 from cuda.core._module import Kernel
@@ -28,4 +26,4 @@ def launch(stream: Stream | GraphBuilder | IsStreamType, config: LaunchConfig, k
         Variable length argument list that is provided to the
         launching kernel.
 
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/_layout.pyi b/cuda_core/cuda/core/_layout.pyi
index 1562a2bf76f..0483a78b6ad 100644
--- a/cuda_core/cuda/core/_layout.pyi
+++ b/cuda_core/cuda/core/_layout.pyi
@@ -1,14 +1,17 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_layout.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_layout.pyx
 
-from __future__ import annotations
+from typing import Any, TypeAlias, TypedDict
 
-import cython
-from libcpp import vector
+OrderFlag: TypeAlias = int
+Property: TypeAlias = int
+extent_t: TypeAlias = int
+stride_t: TypeAlias = int
+axis_t: TypeAlias = int
+axes_mask_t: TypeAlias = int
+property_mask_t: TypeAlias = int
+extents_strides_t: TypeAlias = Any
+axis_vec_t: TypeAlias = Any
 
-OrderFlag = int
-Property = int
-
-@cython.final
 class _StridedLayout:
     """
     A class describing the layout of a multi-dimensional tensor
@@ -39,10 +42,10 @@ class _StridedLayout:
         The offset (as a number of elements, not bytes) of the element at
         index ``(0,) * ndim``. See also :attr:`slice_offset_in_bytes`.
     """
+    itemsize: int
+    slice_offset: stride_t
 
-    def __init__(self: _StridedLayout, shape: tuple[int, ...], strides: tuple[int, ...] | None, itemsize: int, divide_strides: bool=False) -> None:
-        ...
-
+    def __init__(self: _StridedLayout, shape: tuple[int, ...], strides: tuple[int, ...] | None, itemsize: int, divide_strides: bool=False) -> None: ...
     @classmethod
     def dense(cls, shape: tuple[int], itemsize: int, stride_order: str | tuple[int]='C') -> _StridedLayout:
         """
@@ -72,7 +75,6 @@ class _StridedLayout:
             assert _StridedLayout.dense((5, 3, 7), 1, (2, 0, 1)) == _StridedLayout((5, 3, 7), (3, 1, 15), 1)
 
         """
-
     @classmethod
     def dense_like(cls, other: _StridedLayout, stride_order: str | tuple[int]='K') -> _StridedLayout:
         """
@@ -109,13 +111,8 @@ class _StridedLayout:
             assert _StridedLayout.dense_like(layout, "C") == _StridedLayout((7, 5, 3), (15, 3, 1), 1)
             assert _StridedLayout.dense_like(layout, "F") == _StridedLayout((7, 5, 3), (1, 7, 35), 1)
         """
-
-    def __repr__(self: _StridedLayout) -> str:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
+    def __repr__(self: _StridedLayout) -> str: ...
+    def __eq__(self, other: object) -> bool: ...
     @property
     def ndim(self: _StridedLayout) -> int:
         """
@@ -123,7 +120,6 @@ class _StridedLayout:
 
         :type: int
         """
-
     @property
     def shape(self: _StridedLayout) -> tuple[int, ...]:
         """
@@ -131,7 +127,6 @@ class _StridedLayout:
 
         :type: tuple[int]
         """
-
     @property
     def strides(self: _StridedLayout) -> tuple[int, ...] | None:
         """
@@ -141,7 +136,6 @@ class _StridedLayout:
 
         :type: tuple[int] | None
         """
-
     @property
     def strides_in_bytes(self: _StridedLayout) -> tuple[int, ...] | None:
         """
@@ -149,7 +143,6 @@ class _StridedLayout:
 
         :type: tuple[int] | None
         """
-
     @property
     def stride_order(self: _StridedLayout) -> tuple[int, ...]:
         """
@@ -168,7 +161,6 @@ class _StridedLayout:
 
         :type: tuple[int]
         """
-
     @property
     def volume(self: _StridedLayout) -> int:
         """
@@ -176,7 +168,6 @@ class _StridedLayout:
 
         :type: int
         """
-
     @property
     def is_unique(self: _StridedLayout) -> bool:
         """
@@ -196,7 +187,6 @@ class _StridedLayout:
 
         :type: bool
         """
-
     @property
     def is_contiguous_c(self: _StridedLayout) -> bool:
         """
@@ -216,7 +206,6 @@ class _StridedLayout:
 
         :type: bool
         """
-
     @property
     def is_contiguous_f(self: _StridedLayout) -> bool:
         """
@@ -236,7 +225,6 @@ class _StridedLayout:
 
         :type: bool
         """
-
     @property
     def is_contiguous_any(self: _StridedLayout) -> bool:
         """
@@ -275,7 +263,6 @@ class _StridedLayout:
 
         :type: bool
         """
-
     @property
     def is_dense(self: _StridedLayout) -> bool:
         """
@@ -287,7 +274,6 @@ class _StridedLayout:
 
         :type: bool
         """
-
     @property
     def offset_bounds(self: _StridedLayout) -> tuple[int, int]:
         """
@@ -316,7 +302,6 @@ class _StridedLayout:
 
         :type: tuple[int, int]
         """
-
     @property
     def min_offset(self: _StridedLayout) -> int:
         """
@@ -324,7 +309,6 @@ class _StridedLayout:
 
         :type: int
         """
-
     @property
     def max_offset(self: _StridedLayout) -> int:
         """
@@ -332,7 +316,6 @@ class _StridedLayout:
 
         :type: int
         """
-
     @property
     def slice_offset_in_bytes(self: _StridedLayout) -> int:
         """
@@ -345,7 +328,6 @@ class _StridedLayout:
 
         :type: int
         """
-
     def required_size_in_bytes(self: _StridedLayout) -> int:
         """
         The memory allocation size (in bytes) needed so that
@@ -378,13 +360,11 @@ class _StridedLayout:
                 b_view = StridedMemoryView.from_buffer(mem, layout, a_view.dtype)
                 return b_view
         """
-
     def flattened_axis_mask(self: _StridedLayout) -> axes_mask_t:
         """
         A mask describing which axes of this layout are mergeable
         using the :meth:`flattened` method.
         """
-
     def to_dense(self: _StridedLayout, stride_order: object='K') -> _StridedLayout:
         """
         Returns a dense layout with the same shape and itemsize,
@@ -392,7 +372,6 @@ class _StridedLayout:
 
         See :meth:`dense_like` method documentation for details.
         """
-
     def reshaped(self: _StridedLayout, shape: tuple[int]) -> _StridedLayout:
         """
         Returns a layout with the new shape, if the new shape is compatible
@@ -415,13 +394,11 @@ class _StridedLayout:
             assert layout.permuted((2, 0, 1)).reshaped((4, 15,)) == _StridedLayout((4, 15), (1, 4), 1)
             # layout.permuted((2, 0, 1)).reshaped((20, 3)) -> error
         """
-
     def permuted(self: _StridedLayout, axis_order: tuple[int]) -> _StridedLayout:
         """
         Returns a new layout where the shape and strides tuples are permuted
         according to the specified permutation of axes.
         """
-
     def flattened(self: _StridedLayout, start_axis: int=0, end_axis: int=-1, mask: int | None=None) -> _StridedLayout:
         """
         Merges consecutive extents into a single extent (equal to the product of merged extents)
@@ -465,7 +442,6 @@ class _StridedLayout:
             assert layout.flattened(mask=mask) == _StridedLayout((4, 15), (15, 1), 4)
             assert layout2.flattened(mask=mask) == _StridedLayout((4, 15), (1, 4), 4)
         """
-
     def squeezed(self: _StridedLayout) -> _StridedLayout:
         """
         Returns a new layout where all the singleton dimensions (extents equal to 1)
@@ -473,14 +449,12 @@ class _StridedLayout:
         the returned layout will be reduced to a 1-dim layout
         with shape (0,) and strides (0,).
         """
-
     def unsqueezed(self: _StridedLayout, axis: int | tuple[int]) -> _StridedLayout:
         """
         Returns a new layout where the specified axis or axes are added as singleton extents.
         The ``axis`` can be either a single integer in range ``[0, ndim]``
         or a tuple of unique integers in range ``[0, ndim + len(axis) - 1]``.
         """
-
     def broadcast_to(self: _StridedLayout, shape: tuple[int]) -> _StridedLayout:
         """
         Returns a layout with the new shape, if the old shape can be
@@ -494,7 +468,6 @@ class _StridedLayout:
         Strides of the added or modified extents are set to 0, the remaining ones are unchanged.
         If the shapes are not compatible, a ValueError is raised.
         """
-
     def repacked(self: _StridedLayout, itemsize: int, data_ptr: int=0, axis: int=-1, keep_dim: bool=True) -> _StridedLayout:
         """
         Converts the layout to match the specified itemsize.
@@ -546,13 +519,11 @@ class _StridedLayout:
             b = numpy.from_dlpack(complex_view)
             assert b.shape == (5, 3)
         """
-
     def max_compatible_itemsize(self: _StridedLayout, max_itemsize: int=16, data_ptr: int=0, axis: int=-1) -> int:
         """
         Returns the maximum itemsize (but no greater than ``max_itemsize``) that can be used
         with the :meth:`repacked` method for the current layout.
         """
-
     def sliced(self: _StridedLayout, slices: int | slice | tuple[int | slice]) -> _StridedLayout:
         """
         Returns a sliced layout.
@@ -569,13 +540,10 @@ class _StridedLayout:
             any data access.
 
         """
+    def __getitem__(self: _StridedLayout, slices: int | slice | tuple[int | slice]) -> _StridedLayout: ...
 
-    def __getitem__(self: _StridedLayout, slices: int | slice | tuple[int | slice]) -> _StridedLayout:
-        ...
-extent_t = int
-stride_t = int
-axis_t = int
-axes_mask_t = int
-property_mask_t = int
-extents_strides_t = vector.vector
-axis_vec_t = vector.vector
\ No newline at end of file
+class BaseLayout(TypedDict):
+    _mem: extents_strides_t
+    shape: extent_t
+    strides: stride_t
+    ndim: int
diff --git a/cuda_core/cuda/core/_linker.pyi b/cuda_core/cuda/core/_linker.pyi
index 42b08313f78..a649bd69e10 100644
--- a/cuda_core/cuda/core/_linker.pyi
+++ b/cuda_core/cuda/core/_linker.pyi
@@ -1,4 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_linker.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_linker.pyx
 
 """Linking machinery for combining object codes.
 
@@ -6,8 +6,6 @@ This module provides :class:`Linker` for linking one or more
 :class:`~cuda.core.ObjectCode` objects, with :class:`LinkerOptions` for
 configuration.
 """
-from __future__ import annotations
-
 from dataclasses import dataclass
 from typing import Union
 
@@ -16,6 +14,15 @@ import cuda.bindings.nvjitlink
 from cuda.core._module import ObjectCode
 from cuda.core.typing import CompilerBackendType, ObjectCodeFormatType
 
+_keep_driver_in_stub: cuda.bindings.driver.CUlinkState
+_keep_nvjitlink_in_stub: cuda.bindings.nvjitlink.nvJitLinkHandle
+__all__ = ['Linker', 'LinkerOptions']
+LinkerHandleT = Union['cuda.bindings.nvjitlink.nvJitLinkHandle', 'cuda.bindings.driver.CUlinkState']
+_driver = None
+_inited = False
+_use_nvjitlink_backend = None
+_nvjitlink_input_types = None
+_driver_input_types = None
 
 class Linker:
     """Represent a linking machinery to link one or more object codes into
@@ -31,10 +38,7 @@ class Linker:
     options : :class:`LinkerOptions`, optional
         Options for the linker. If not provided, default options will be used.
     """
-
-    def __init__(self, options: LinkerOptions | None=None, *object_codes: ObjectCode):
-        ...
-
+    def __init__(self, *object_codes: ObjectCode, options: LinkerOptions | None=None): ...
     def link(self, target_type: ObjectCodeFormatType | str) -> ObjectCode:
         """Link the provided object codes into a single output of the specified target type.
 
@@ -53,7 +57,6 @@ class Linker:
             Ensure that input object codes were compiled with appropriate
             flags for linking (e.g., relocatable device code enabled).
         """
-
     def get_error_log(self) -> str:
         """Get the error log generated by the linker.
 
@@ -62,7 +65,6 @@ class Linker:
         str
             The error log.
         """
-
     def get_info_log(self) -> str:
         """Get the info log generated by the linker.
 
@@ -71,10 +73,8 @@ class Linker:
         str
             The info log.
         """
-
     def close(self) -> None:
         """Destroy this linker."""
-
     @property
     def handle(self) -> LinkerHandleT:
         """Return the underlying handle object.
@@ -88,7 +88,6 @@ class Linker:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(Linker.handle)``.
         """
-
     @classmethod
     def which_backend(cls) -> CompilerBackendType:
         """Return which linking backend will be used.
@@ -201,15 +200,9 @@ class LinkerOptions:
     no_cache: bool | None = None
     numba_debug: bool | None = None
 
-    def __post_init__(self) -> None:
-        ...
-
-    def _prepare_nvjitlink_options(self, as_bytes: bool=False) -> list[bytes] | list[str]:
-        ...
-
-    def _prepare_driver_options(self) -> tuple[list[object], list[object]]:
-        ...
-
+    def __post_init__(self) -> None: ...
+    def _prepare_nvjitlink_options(self, as_bytes: bool=False) -> list[bytes] | list[str]: ...
+    def _prepare_driver_options(self) -> tuple[list[object], list[object]]: ...
     def as_bytes(self, backend: str='nvjitlink') -> list[bytes]:
         """Convert linker options to bytes format for the nvjitlink backend.
 
@@ -230,21 +223,8 @@ class LinkerOptions:
         RuntimeError
             If nvJitLink backend is not available.
         """
-_keep_driver_in_stub: 'cuda.bindings.driver.CUlinkState'
-_keep_nvjitlink_in_stub: 'cuda.bindings.nvjitlink.nvJitLinkHandle'
-__all__ = ['Linker', 'LinkerOptions']
-LinkerHandleT = Union['cuda.bindings.nvjitlink.nvJitLinkHandle', 'cuda.bindings.driver.CUlinkState']
-_driver = None
-_inited = False
-_use_nvjitlink_backend = None
-_nvjitlink_input_types = None
-_driver_input_types = None
-
-def _nvjitlink_has_version_symbol(nvjitlink) -> bool:
-    ...
 
+def _nvjitlink_has_version_symbol(nvjitlink) -> bool: ...
 def _decide_nvjitlink_or_driver() -> bool:
     """Return True if falling back to the cuLink* driver APIs."""
-
-def _lazy_init() -> None:
-    ...
\ No newline at end of file
+def _lazy_init() -> None: ...
diff --git a/cuda_core/cuda/core/_memory/_buffer.pyi b/cuda_core/cuda/core/_memory/_buffer.pyi
index b441754503c..f0ba484217e 100644
--- a/cuda_core/cuda/core/_memory/_buffer.pyi
+++ b/cuda_core/cuda/core/_memory/_buffer.pyi
@@ -1,8 +1,7 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_buffer.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_buffer.pyx
 
-from __future__ import annotations
+from typing import TypedDict
 
-import cython
 from cuda.core._memory._copy_enums import CopyOptions
 from cuda.core._memory._device_memory_resource import DeviceMemoryResource
 from cuda.core._memory._ipc import IPCBufferDescriptor
@@ -12,6 +11,7 @@ from cuda.core._utils.pycompat import BufferProtocol
 from cuda.core.graph import GraphBuilder
 from cuda.core.typing import DevicePointerType
 
+__all__ = ['Buffer', 'MemoryResource']
 
 class Buffer:
     """Represent a handle to allocated memory.
@@ -29,16 +29,10 @@ class Buffer:
     by calling :meth:`from_ipc_descriptor` and therefore performs an IPC
     import. Do not unpickle buffers from untrusted sources.
     """
+    _size: int
 
-    def __cinit__(self) -> None:
-        ...
-
-    def _clear(self) -> None:
-        ...
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def _clear(self) -> None: ...
+    def __init__(self, *args, **kwargs) -> None: ...
     @classmethod
     def _init(cls, ptr: DevicePointerType, size: int, mr: MemoryResource | None=None, ipc_descriptor: IPCBufferDescriptor | None=None, owner: object | None=None, *, stream: Stream | GraphBuilder | None=None) -> Buffer:
         """Create a Buffer from a raw pointer.
@@ -50,14 +44,9 @@ class Buffer:
         (``stream`` if given, otherwise ``default_stream()``). Recording a
         default-stream token requires a CUDA context to be current.
         """
-
     @staticmethod
-    def _reduce_helper(mr, ipc_descriptor):
-        ...
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-
+    def _reduce_helper(mr, ipc_descriptor): ...
+    def __reduce__(self) -> tuple[object, ...]: ...
     @staticmethod
     def from_handle(ptr: DevicePointerType, size: int, mr: MemoryResource | None=None, owner: object | None=None, *, stream: Stream | GraphBuilder | None=None) -> Buffer:
         """Create a new :class:`Buffer` object from a pointer.
@@ -90,7 +79,6 @@ class Buffer:
         non-owning reference.  The pointer will NOT be freed when the
         :class:`Buffer` is closed or garbage collected.
         """
-
     @classmethod
     def from_ipc_descriptor(cls, mr: DeviceMemoryResource | PinnedMemoryResource, ipc_descriptor: IPCBufferDescriptor, *, stream: Stream) -> Buffer:
         """Import a buffer that was exported from another process.
@@ -111,12 +99,9 @@ class Buffer:
         and must be treated as untrusted input unless the peer is known to be
         cooperating.
         """
-
     @property
-    @cython.critical_section
     def ipc_descriptor(self) -> IPCBufferDescriptor:
         """Descriptor for sharing this buffer with other processes."""
-
     def close(self, stream: Stream | GraphBuilder | None=None) -> None:
         """Deallocate this buffer asynchronously on the given stream.
 
@@ -134,7 +119,6 @@ class Buffer:
         set_deallocation_stream
             Change the deallocation stream without closing the buffer.
         """
-
     def set_deallocation_stream(self, stream: Stream | GraphBuilder) -> None:
         """Change the stream that orders this buffer's eventual deallocation.
 
@@ -164,13 +148,8 @@ class Buffer:
         Synchronizing concurrent mutation and destruction of the same buffer
         is the caller's responsibility.
         """
-
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc_val, exc_tb):
-        ...
-
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc_val, exc_tb): ...
     def copy_to(self, dst: Buffer | None=None, *, stream: Stream | GraphBuilder, options: CopyOptions | None=None) -> Buffer:
         """Copy from this buffer to the dst buffer asynchronously on the given stream.
 
@@ -210,7 +189,6 @@ class Buffer:
             back to a plain copy cannot honor that guarantee.
 
         """
-
     def copy_from(self, src: Buffer, *, stream: Stream | GraphBuilder, options: CopyOptions | None=None) -> None:
         """Copy from the src buffer to this buffer asynchronously on the given stream.
 
@@ -244,7 +222,6 @@ class Buffer:
             cuda.bindings or the driver is older than CUDA 13.2: falling
             back to a plain copy cannot honor that guarantee.
         """
-
     def fill(self, value: int | BufferProtocol, *, stream: Stream | GraphBuilder) -> None:
         """Fill this buffer with a repeating byte pattern.
 
@@ -267,23 +244,13 @@ class Buffer:
             If int value is outside [0, 256).
 
         """
-
-    def __dlpack__(self, *, stream: int | None=None, max_version: tuple[int, int] | None=None, dl_device: tuple[int, int] | None=None, copy: bool | None=None) -> object:
-        ...
-
-    def __dlpack_device__(self) -> tuple[int, int]:
-        ...
-
-    def __buffer__(self, flags: int, /) -> memoryview:
-        ...
-
-    def __release_buffer__(self, buffer: memoryview, /) -> None:
-        ...
-
+    def __dlpack__(self, *, stream: int | None=None, max_version: tuple[int, int] | None=None, dl_device: tuple[int, int] | None=None, copy: bool | None=None) -> object: ...
+    def __dlpack_device__(self) -> tuple[int, int]: ...
+    def __buffer__(self, flags: int, /) -> memoryview: ...
+    def __release_buffer__(self, buffer: memoryview, /) -> None: ...
     @property
     def device_id(self) -> int:
         """Return the device ordinal of this buffer."""
-
     @property
     def handle(self) -> int:
         """Return the buffer handle object.
@@ -293,40 +260,27 @@ class Buffer:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(Buffer.handle)``.
         """
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
+    def __repr__(self) -> str: ...
     @property
     def is_device_accessible(self) -> bool:
         """Return True if this buffer can be accessed by the GPU, otherwise False."""
-
     @property
     def is_host_accessible(self) -> bool:
         """Return True if this buffer can be accessed by the CPU, otherwise False."""
-
     @property
     def is_managed(self) -> bool:
         """Return True if this buffer is CUDA managed (unified) memory, otherwise False."""
-
     @property
     def is_mapped(self) -> bool:
         """Return True if this buffer is mapped into the process via IPC."""
-
     @property
     def memory_resource(self) -> MemoryResource:
         """Return the memory resource associated with this buffer."""
-
     @property
     def size(self) -> int:
         """Return the memory size of this buffer."""
-
     @property
     def owner(self) -> object:
         """Return the object holding external allocation."""
@@ -342,7 +296,6 @@ class MemoryResource:
     buffer properties are retrieved simply by looking up the underlying memory
     resource's respective property.)
     """
-
     def allocate(self, size: int, *, stream: Stream | GraphBuilder) -> Buffer:
         """Allocate a buffer of the requested size.
 
@@ -366,7 +319,6 @@ class MemoryResource:
             The allocated buffer object, which can be used for device or host operations
             depending on the resource's properties.
         """
-
     def deallocate(self, ptr: DevicePointerType, size: int, *, stream: Stream | GraphBuilder) -> None:
         """Deallocate a buffer previously allocated by this resource.
 
@@ -381,20 +333,21 @@ class MemoryResource:
             asynchronously. Must be passed explicitly; pass
             ``device.default_stream`` to use the default stream.
         """
-
     @property
     def is_device_accessible(self) -> bool:
         """Whether buffers allocated by this resource are device-accessible."""
-
     @property
     def is_host_accessible(self) -> bool:
         """Whether buffers allocated by this resource are host-accessible."""
-
     @property
     def is_managed(self) -> bool:
         """Whether buffers allocated by this resource are CUDA managed (unified) memory."""
-
     @property
     def device_id(self) -> int:
         """Device ID associated with this memory resource, or -1 if not applicable."""
-__all__ = ['Buffer', 'MemoryResource']
\ No newline at end of file
+
+class _MemAttrs(TypedDict):
+    device_id: int
+    is_device_accessible: bool
+    is_host_accessible: bool
+    is_managed: bool
diff --git a/cuda_core/cuda/core/_memory/_copy_attributes.pyi b/cuda_core/cuda/core/_memory/_copy_attributes.pyi
index 0fceb058f53..93b9a97aa2b 100644
--- a/cuda_core/cuda/core/_memory/_copy_attributes.pyi
+++ b/cuda_core/cuda/core/_memory/_copy_attributes.pyi
@@ -1,3 +1 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_copy_attributes.pyx
-
-from __future__ import annotations
\ No newline at end of file
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_copy_attributes.pyx
diff --git a/cuda_core/cuda/core/_memory/_copy_ops.pyi b/cuda_core/cuda/core/_memory/_copy_ops.pyi
index 9923f2a0042..f3439732230 100644
--- a/cuda_core/cuda/core/_memory/_copy_ops.pyi
+++ b/cuda_core/cuda/core/_memory/_copy_ops.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_copy_ops.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_copy_ops.pyx
 
 from collections.abc import Sequence
 
@@ -21,7 +19,6 @@ def _normalize_copy_options(options: CopyOptions | Sequence[CopyOptions] | None,
     this expansion (and the run encoding applied to it) is the only
     observable evidence that a scalar reached every copy.
     """
-
 def copy_batch(stream: Stream, srcs: Sequence[Buffer], dsts: Sequence[Buffer], *, options: CopyOptions | Sequence[CopyOptions] | None=None) -> None:
     """Copy a batch of buffers asynchronously.
 
@@ -93,4 +90,4 @@ def copy_batch(stream: Stream, srcs: Sequence[Buffer], dsts: Sequence[Buffer], *
     stream-ordered access would let a caller reuse the source buffer before
     the real read happens.
 
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/_memory/_device_memory_resource.pyi b/cuda_core/cuda/core/_memory/_device_memory_resource.pyi
index efe845377c0..897e1d03302 100644
--- a/cuda_core/cuda/core/_memory/_device_memory_resource.pyi
+++ b/cuda_core/cuda/core/_memory/_device_memory_resource.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_device_memory_resource.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_device_memory_resource.pyx
 
 import uuid
 from dataclasses import dataclass
@@ -10,6 +8,7 @@ from cuda.core._memory._ipc import IPCAllocationHandle
 from cuda.core._memory._memory_pool import _MemPool
 from cuda.core._memory._peer_access_utils import PeerAccessibleBySetProxy
 
+__all__ = ['DeviceMemoryResource', 'DeviceMemoryResourceOptions']
 
 @dataclass
 class DeviceMemoryResourceOptions:
@@ -116,16 +115,8 @@ class DeviceMemoryResource(_MemPool):
     descriptors from trusted peers, and do not unpickle buffers from
     untrusted sources.
     """
-
-    def __cinit__(self, *args, **kwargs) -> None:
-        ...
-
-    def __init__(self, device_id: Device | int, options: DeviceMemoryResourceOptions | None=None) -> None:
-        ...
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-
+    def __init__(self, device_id: Device | int, options: DeviceMemoryResourceOptions | None=None) -> None: ...
+    def __reduce__(self) -> tuple[object, ...]: ...
     @staticmethod
     def from_registry(uuid: uuid.UUID) -> DeviceMemoryResource:
         """
@@ -136,7 +127,6 @@ class DeviceMemoryResource(_MemPool):
         RuntimeError
             If no mapped memory resource is found in the registry.
         """
-
     def register(self, uuid: uuid.UUID) -> DeviceMemoryResource:
         """
         Register a mapped memory resource.
@@ -146,7 +136,6 @@ class DeviceMemoryResource(_MemPool):
         The registered mapped memory resource. If one was previously registered
         with the given key, it is returned.
         """
-
     @classmethod
     def from_allocation_handle(cls, device_id: Device | int, alloc_handle: int | IPCAllocationHandle) -> DeviceMemoryResource:
         """Create a device memory resource from an allocation handle.
@@ -170,7 +159,6 @@ class DeviceMemoryResource(_MemPool):
         -------
             A new device memory resource instance with the imported handle.
         """
-
     @property
     def allocation_handle(self) -> IPCAllocationHandle:
         """Shareable handle for this memory pool (requires IPC).
@@ -178,11 +166,9 @@ class DeviceMemoryResource(_MemPool):
         The handle can be used to share the memory pool with other processes.
         The handle is cached in this `MemoryResource` and owned by it.
         """
-
     @property
     def device_id(self) -> int:
         """The associated device ordinal."""
-
     @property
     def peer_accessible_by(self) -> PeerAccessibleBySetProxy:
         """
@@ -202,19 +188,14 @@ class DeviceMemoryResource(_MemPool):
         >>> dmr.peer_accessible_by.add(2)  # update access to include device 2
         >>> dmr.peer_accessible_by = []    # revoke peer access
         """
-
     @peer_accessible_by.setter
-    def peer_accessible_by(self, devices) -> None:
-        ...
-
+    def peer_accessible_by(self, devices) -> None: ...
     @property
     def is_device_accessible(self) -> bool:
         """Return True. This memory resource provides device-accessible buffers."""
-
     @property
     def is_host_accessible(self) -> bool:
         """Return False. This memory resource does not provide host-accessible buffers."""
-__all__ = ['DeviceMemoryResource', 'DeviceMemoryResourceOptions']
 
 def DMR_mempool_get_access(dmr: DeviceMemoryResource, device_id: int) -> str:
     """
@@ -230,6 +211,4 @@ def DMR_mempool_get_access(dmr: DeviceMemoryResource, device_id: int) -> str:
     str
         Access permissions: "rw" for read-write, "r" for read-only, "" for no access.
     """
-
-def _deep_reduce_device_memory_resource(mr) -> tuple[object, ...]:
-    ...
\ No newline at end of file
+def _deep_reduce_device_memory_resource(mr) -> tuple[object, ...]: ...
diff --git a/cuda_core/cuda/core/_memory/_graph_memory_resource.pyi b/cuda_core/cuda/core/_memory/_graph_memory_resource.pyi
index b34f968fdc9..5801490f8fb 100644
--- a/cuda_core/cuda/core/_memory/_graph_memory_resource.pyi
+++ b/cuda_core/cuda/core/_memory/_graph_memory_resource.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_graph_memory_resource.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_graph_memory_resource.pyx
 
 from cuda.core._device import Device
 from cuda.core._memory._buffer import Buffer, MemoryResource
@@ -8,82 +6,59 @@ from cuda.core._stream import Stream
 from cuda.core.graph import GraphBuilder
 from cuda.core.typing import DevicePointerType
 
+__all__ = ['GraphMemoryResource']
 
 class GraphMemoryResourceAttributes:
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @classmethod
-    def _init(cls, device_id: int) -> GraphMemoryResourceAttributes:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-
+    def _init(cls, device_id: int) -> GraphMemoryResourceAttributes: ...
+    def __repr__(self) -> str: ...
     @property
     def reserved_mem_current(self) -> int:
         """Current amount of backing memory allocated."""
-
     @property
     def reserved_mem_high(self) -> int:
         """
         High watermark of backing memory allocated. It can be set to zero to
         reset it to the current usage.
         """
-
     @reserved_mem_high.setter
-    def reserved_mem_high(self, value: int) -> None:
-        ...
-
+    def reserved_mem_high(self, value: int) -> None: ...
     @property
     def used_mem_current(self) -> int:
         """Current amount of memory in use."""
-
     @property
     def used_mem_high(self) -> int:
         """
         High watermark of memory in use. It can be set to zero to reset it to
         the current usage.
         """
-
     @used_mem_high.setter
-    def used_mem_high(self, value: int) -> None:
-        ...
+    def used_mem_high(self, value: int) -> None: ...
 
 class cyGraphMemoryResource(MemoryResource):
-
-    def __cinit__(self, device_id: int) -> None:
-        ...
-
+    def __init__(self, device_id: int) -> None: ...
     def allocate(self, size: int, *, stream: Stream | GraphBuilder) -> Buffer:
         """
         Allocate a buffer of the requested size. See documentation for :obj:`~_memory.MemoryResource`.
         """
-
     def deallocate(self, ptr: DevicePointerType, size: int, *, stream: Stream | GraphBuilder) -> None:
         """
         Deallocate a buffer of the requested size. See documentation for :obj:`~_memory.MemoryResource`.
         """
-
     def close(self) -> None:
         """No operation (provided for compatibility)."""
-
     def trim(self) -> None:
         """Free unused memory that was cached on the specified device for use with graphs back to the OS."""
-
     @property
     def attributes(self) -> GraphMemoryResourceAttributes:
         """Asynchronous allocation attributes related to graphs."""
-
     @property
     def device_id(self) -> int:
         """The associated device ordinal."""
-
     @property
     def is_device_accessible(self) -> bool:
         """Return True. This memory resource provides device-accessible buffers."""
-
     @property
     def is_host_accessible(self) -> bool:
         """Return False. This memory resource does not provide host-accessible buffers."""
@@ -106,11 +81,6 @@ class GraphMemoryResource(cyGraphMemoryResource):
     device_id: int | Device
         Device or Device ordinal for which a graph memory resource is obtained.
     """
-
-    def __new__(cls, device_id: int | Device) -> GraphMemoryResource:
-        ...
-
+    def __new__(cls, device_id: int | Device) -> GraphMemoryResource: ...
     @classmethod
-    def _create(cls, device_id: int) -> GraphMemoryResource:
-        ...
-__all__ = ['GraphMemoryResource']
\ No newline at end of file
+    def _create(cls, device_id: int) -> GraphMemoryResource: ...
diff --git a/cuda_core/cuda/core/_memory/_ipc.pyi b/cuda_core/cuda/core/_memory/_ipc.pyi
index 7c707ab0418..d1b2324b8f4 100644
--- a/cuda_core/cuda/core/_memory/_ipc.pyi
+++ b/cuda_core/cuda/core/_memory/_ipc.pyi
@@ -1,41 +1,26 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_ipc.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_ipc.pyx
 
 import uuid
 
+__all__ = []
 
 class IPCDataForBuffer:
     """Data members related to sharing memory buffers via IPC."""
-
-    def __cinit__(self, ipc_descriptor: IPCBufferDescriptor, is_mapped: bool) -> None:
-        ...
-
+    def __init__(self, ipc_descriptor: IPCBufferDescriptor, is_mapped: bool) -> None: ...
     @property
-    def ipc_descriptor(self) -> IPCBufferDescriptor:
-        ...
-
+    def ipc_descriptor(self) -> IPCBufferDescriptor: ...
     @property
-    def is_mapped(self) -> bool:
-        ...
+    def is_mapped(self) -> bool: ...
 
 class IPCDataForMR:
     """Data members related to sharing memory resources via IPC."""
-
-    def __cinit__(self, alloc_handle: IPCAllocationHandle, is_mapped: bool) -> None:
-        ...
-
+    def __init__(self, alloc_handle: IPCAllocationHandle, is_mapped: bool) -> None: ...
     @property
-    def alloc_handle(self) -> IPCAllocationHandle:
-        ...
-
+    def alloc_handle(self) -> IPCAllocationHandle: ...
     @property
-    def is_mapped(self) -> bool:
-        ...
-
+    def is_mapped(self) -> bool: ...
     @property
-    def uuid(self) -> uuid.UUID | None:
-        ...
+    def uuid(self) -> uuid.UUID | None: ...
 
 class IPCBufferDescriptor:
     """Serializable object describing a buffer that can be shared between processes.
@@ -46,48 +31,25 @@ class IPCBufferDescriptor:
     Receivers must treat them as untrusted and import only through
     :meth:`Buffer.from_ipc_descriptor`.
     """
-
-    def __init__(self, *arg, **kwargs) -> None:
-        ...
-
+    def __init__(self, *arg, **kwargs) -> None: ...
     @staticmethod
-    def _init(reserved: bytes, size: int) -> IPCBufferDescriptor:
-        ...
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-
+    def _init(reserved: bytes, size: int) -> IPCBufferDescriptor: ...
+    def __reduce__(self) -> tuple[object, ...]: ...
     @property
-    def size(self) -> int:
-        ...
+    def size(self) -> int: ...
 
 class IPCAllocationHandle:
     """Shareable handle to an IPC-enabled device memory pool."""
-
+    def __init__(self, *arg, **kwargs) -> None: ...
+    @classmethod
+    def _init(cls, handle: int, uuid: uuid.UUID | None) -> IPCAllocationHandle: ...
     def close(self):
         """Close the handle."""
-
-    def __init__(self, *arg, **kwargs) -> None:
-        ...
-
-    @classmethod
-    def _init(cls, handle: int, uuid: uuid.UUID | None) -> IPCAllocationHandle:
-        ...
-
-    def __int__(self) -> int:
-        ...
-
+    def __int__(self) -> int: ...
     @property
-    def handle(self) -> int:
-        ...
-
+    def handle(self) -> int: ...
     @property
-    def uuid(self) -> uuid.UUID:
-        ...
-__all__ = []
-
-def _reduce_allocation_handle(alloc_handle: IPCAllocationHandle) -> tuple[object, ...]:
-    ...
+    def uuid(self) -> uuid.UUID: ...
 
-def _reconstruct_allocation_handle(cls: type, df: object, uuid: uuid.UUID | None) -> IPCAllocationHandle:
-    ...
\ No newline at end of file
+def _reduce_allocation_handle(alloc_handle: IPCAllocationHandle) -> tuple[object, ...]: ...
+def _reconstruct_allocation_handle(cls: type, df: object, uuid: uuid.UUID | None) -> IPCAllocationHandle: ...
diff --git a/cuda_core/cuda/core/_memory/_location.pyi b/cuda_core/cuda/core/_memory/_location.pyi
new file mode 100644
index 00000000000..fa1467c3432
--- /dev/null
+++ b/cuda_core/cuda/core/_memory/_location.pyi
@@ -0,0 +1 @@
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_location.pxd
diff --git a/cuda_core/cuda/core/_memory/_managed_memory_ops.pyi b/cuda_core/cuda/core/_memory/_managed_memory_ops.pyi
index a72bc52827f..4523a39d728 100644
--- a/cuda_core/cuda/core/_memory/_managed_memory_ops.pyi
+++ b/cuda_core/cuda/core/_memory/_managed_memory_ops.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_managed_memory_ops.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_managed_memory_ops.pyx
 
 from collections.abc import Sequence
 
@@ -34,17 +32,14 @@ def discard_batch(stream: Stream | GraphBuilder, buffers: Sequence[Buffer]) -> N
     NotImplementedError
         On a CUDA 12 build of ``cuda.core``.
     """
-
 def _do_single_discard_py(buf: Buffer, stream: Stream | GraphBuilder | None) -> None:
     """Internal: single-buffer discard for ManagedBuffer.discard()."""
-
 def _advise_one(buf: Buffer, advice: driver.CUmem_advise, location: Device | Host | None) -> None:
     """Internal: apply managed-memory advice to a single buffer.
 
     Used by :class:`ManagedBuffer` property setters. Not part of the
     public API.
     """
-
 def prefetch_batch(stream: Stream | GraphBuilder, buffers: Sequence[Buffer], locations: Device | Host | Sequence[Device | Host]) -> None:
     """Prefetch a batch of managed-memory ranges to target locations.
 
@@ -68,16 +63,12 @@ def prefetch_batch(stream: Stream | GraphBuilder, buffers: Sequence[Buffer], loc
     ``cuMemPrefetchAsync`` per buffer (no batched driver entry point on
     CUDA 12). CUDA 13 builds use ``cuMemPrefetchBatchAsync`` directly.
     """
-
 def _do_single_prefetch_py(buf: Buffer, location: Device | Host | None, stream: Stream | GraphBuilder | None) -> None:
     """Internal: single-buffer prefetch for ManagedBuffer.prefetch().
 
     Uses cuMemPrefetchAsync (works on CUDA 12 and 13).
     """
-
-def _read_preferred_location_v2(buf: Buffer) -> Device | Host | None:
-    ...
-
+def _read_preferred_location_v2(buf: Buffer) -> Device | Host | None: ...
 def discard_prefetch_batch(stream: Stream | GraphBuilder, buffers: Sequence[Buffer], locations: Device | Host | Sequence[Device | Host]) -> None:
     """Discard a batch of managed-memory ranges and prefetch them to target locations.
 
@@ -100,7 +91,6 @@ def discard_prefetch_batch(stream: Stream | GraphBuilder, buffers: Sequence[Buff
     NotImplementedError
         On a CUDA 12 build of ``cuda.core``.
     """
-
 def _do_single_discard_prefetch_py(buf: Buffer, location: Device | Host | None, stream: Stream | GraphBuilder | None) -> None:
     """Internal: single-buffer discard+prefetch for
-    ManagedBuffer.discard_prefetch()."""
\ No newline at end of file
+    ManagedBuffer.discard_prefetch()."""
diff --git a/cuda_core/cuda/core/_memory/_managed_memory_resource.pyi b/cuda_core/cuda/core/_memory/_managed_memory_resource.pyi
index 511c72252b4..44523ae0f53 100644
--- a/cuda_core/cuda/core/_memory/_managed_memory_resource.pyi
+++ b/cuda_core/cuda/core/_memory/_managed_memory_resource.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_managed_memory_resource.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_managed_memory_resource.pyx
 
 from dataclasses import dataclass
 
@@ -10,6 +8,7 @@ from cuda.core._stream import Stream
 from cuda.core.graph import GraphBuilder
 from cuda.core.typing import ManagedMemoryLocationType
 
+__all__ = ['ManagedMemoryResource', 'ManagedMemoryResourceOptions']
 
 @dataclass
 class ManagedMemoryResourceOptions:
@@ -76,10 +75,7 @@ class ManagedMemoryResource(_MemPool):
     IPC (Inter-Process Communication) is not currently supported for managed
     memory pools.
     """
-
-    def __init__(self, options: ManagedMemoryResourceOptions | None=None) -> None:
-        ...
-
+    def __init__(self, options: ManagedMemoryResourceOptions | None=None) -> None: ...
     def allocate(self, size: int, *, stream: Stream | GraphBuilder) -> ManagedBuffer:
         """Allocate a managed-memory buffer of the requested size.
 
@@ -101,11 +97,9 @@ class ManagedMemoryResource(_MemPool):
             and instance methods (``prefetch``, ``discard``,
             ``discard_prefetch``).
         """
-
     @property
     def device_id(self) -> int:
         """The preferred device ordinal, or -1 if the preferred location is not a device."""
-
     @property
     def preferred_location(self) -> tuple[ManagedMemoryLocationType, int | None] | None:
         """The preferred location for managed memory allocations.
@@ -115,19 +109,15 @@ class ManagedMemoryResource(_MemPool):
         ``"host"``, or ``"host_numa"``, and *id* is the device ordinal,
         ``None`` (for ``"host"``), or the NUMA node ID, respectively.
         """
-
     @property
     def is_device_accessible(self) -> bool:
         """Return True. This memory resource provides device-accessible buffers."""
-
     @property
     def is_host_accessible(self) -> bool:
         """Return True. This memory resource provides host-accessible buffers."""
-
     @property
     def is_managed(self) -> bool:
         """Return True. This memory resource provides managed (unified) memory buffers."""
-__all__ = ['ManagedMemoryResource', 'ManagedMemoryResourceOptions']
 
 def reset_concurrent_access_warning() -> None:
-    """Reset the concurrent access warning flag for testing purposes."""
\ No newline at end of file
+    """Reset the concurrent access warning flag for testing purposes."""
diff --git a/cuda_core/cuda/core/_memory/_memory_pool.pyi b/cuda_core/cuda/core/_memory/_memory_pool.pyi
index 7f8c64aedda..70189ee5413 100644
--- a/cuda_core/cuda/core/_memory/_memory_pool.pyi
+++ b/cuda_core/cuda/core/_memory/_memory_pool.pyi
@@ -1,10 +1,7 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_memory_pool.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_memory_pool.pyx
 
 import uuid
 
-import cython
 from cuda.core._memory._buffer import Buffer, MemoryResource
 from cuda.core._stream import Stream
 from cuda.core.graph import GraphBuilder
@@ -13,56 +10,40 @@ from cuda.core.typing import DevicePointerType
 
 class _MemPoolAttributes:
     """Provides access to memory pool attributes."""
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
+    def __repr__(self) -> str: ...
     @property
     def reuse_follow_event_dependencies(self) -> bool:
         """Allow memory to be reused when there are event dependencies between streams."""
-
     @property
     def reuse_allow_opportunistic(self) -> bool:
         """Allow reuse of completed frees without dependencies."""
-
     @property
     def reuse_allow_internal_dependencies(self) -> bool:
         """Allow insertion of new stream dependencies for memory reuse."""
-
     @property
     def release_threshold(self) -> int:
         """Amount of reserved memory to hold before OS release."""
-
     @property
     def reserved_mem_current(self) -> int:
         """Current amount of backing memory allocated."""
-
     @property
     def reserved_mem_high(self) -> int:
         """High watermark of backing memory allocated."""
-
     @property
     def used_mem_current(self) -> int:
         """Current amount of memory in use."""
-
     @property
     def used_mem_high(self) -> int:
         """High watermark of memory in use."""
 
 class _MemPool(MemoryResource):
-
-    def __cinit__(self) -> None:
-        ...
-
+    def __init__(self) -> None: ...
     def close(self) -> None:
         """
         Close the memory resource and destroy the associated memory pool
         if owned.
         """
-
     def allocate(self, size: int, *, stream: Stream | GraphBuilder) -> Buffer:
         """Allocate a buffer of the requested size.
 
@@ -81,7 +62,6 @@ class _MemPool(MemoryResource):
             The allocated buffer object, which is accessible on the device that this memory
             resource was created for.
         """
-
     def deallocate(self, ptr: DevicePointerType, size: int, *, stream: Stream | GraphBuilder) -> None:
         """Deallocate a buffer previously allocated by this resource.
 
@@ -96,34 +76,27 @@ class _MemPool(MemoryResource):
             asynchronously. Must be passed explicitly; pass
             ``device.default_stream`` to use the default stream.
         """
-
     @property
-    @cython.critical_section
     def attributes(self) -> _MemPoolAttributes:
         """Memory pool attributes."""
-
     @property
     def handle(self) -> object:
         """Handle to the underlying memory pool."""
-
     @property
     def is_handle_owned(self) -> bool:
         """Whether the memory resource handle is owned. If False, ``close`` has no effect."""
-
     @property
     def is_ipc_enabled(self) -> bool:
         """Whether this memory resource has IPC enabled."""
-
     @property
     def is_mapped(self) -> bool:
         """
         Whether this is a mapping of an IPC-enabled memory resource from
         another process.  If True, allocation is not permitted.
         """
-
     @property
     def uuid(self) -> uuid.UUID | None:
         """
         A universally unique identifier for this memory resource. Meaningful
         only for IPC-enabled memory resources.
-        """
\ No newline at end of file
+        """
diff --git a/cuda_core/cuda/core/_memory/_peer_access_utils.pyi b/cuda_core/cuda/core/_memory/_peer_access_utils.pyi
index fa2b5c490a4..2c38debd2de 100644
--- a/cuda_core/cuda/core/_memory/_peer_access_utils.pyi
+++ b/cuda_core/cuda/core/_memory/_peer_access_utils.pyi
@@ -1,10 +1,8 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_peer_access_utils.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_peer_access_utils.pyx
 
-from __future__ import annotations
-
-from collections.abc import Callable, Iterable, Iterator, Set
+from collections.abc import Iterable, Iterator, Set
 from dataclasses import dataclass
-from typing import Any
+from typing import Any, Callable
 
 from cuda.core._device import Device
 from cuda.core._memory._device_memory_resource import DeviceMemoryResource
@@ -35,60 +33,33 @@ class PeerAccessibleBySetProxy:
     is updated, so coalescing into a single driver call lets the toolkit handle
     the mappings in parallel.
     """
-    __slots__ = ('_mr',)
-
-    def __init__(self, mr: DeviceMemoryResource) -> None:
-        ...
+    __slots__ = '_mr'
 
+    def __init__(self, mr: DeviceMemoryResource) -> None: ...
     @classmethod
-    def _from_iterable(cls, it: Iterable[Device]) -> set[Device]:
-        ...
-
-    def __contains__(self, value: object) -> bool:
-        ...
-
-    def __iter__(self) -> Iterator[Device]:
-        ...
-
-    def __len__(self) -> int:
-        ...
-
+    def _from_iterable(cls, it: Iterable[Device]) -> set[Device]: ...
+    def __contains__(self, value: object) -> bool: ...
+    def __iter__(self) -> Iterator[Device]: ...
+    def __len__(self) -> int: ...
     def add(self, value: Device | int) -> None:
         """Grant peer access from ``value`` to allocations in this pool."""
-
     def discard(self, value: Device | int) -> None:
         """Revoke peer access from ``value`` to allocations in this pool."""
-
     def clear(self) -> None:
         """Revoke all peer access in a single driver call."""
-
     def update(self, *others: Iterable[Device | int]) -> None:
         """Grant peer access to every device in ``others`` in one driver call."""
-
     def difference_update(self, *others: Iterable[Device | int]) -> None:
         """Revoke peer access for every device in ``others`` in one driver call."""
-
     def intersection_update(self, *others: Iterable[Device | int]) -> None:
         """Restrict peer access to the intersection in a single driver call."""
-
     def symmetric_difference_update(self, other: Iterable[Device | int]) -> None:
         """Toggle peer access for every device in ``other`` in one driver call."""
-
-    def __ior__(self, other: Set[Any]) -> PeerAccessibleBySetProxy:
-        ...
-
-    def __iand__(self, other: Set[Any]) -> PeerAccessibleBySetProxy:
-        ...
-
-    def __isub__(self, other: Set[Any]) -> PeerAccessibleBySetProxy:
-        ...
-
-    def __ixor__(self, other: Set[Any]) -> PeerAccessibleBySetProxy:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-
+    def __ior__(self, other: Set[Any]) -> PeerAccessibleBySetProxy: ...
+    def __iand__(self, other: Set[Any]) -> PeerAccessibleBySetProxy: ...
+    def __isub__(self, other: Set[Any]) -> PeerAccessibleBySetProxy: ...
+    def __ixor__(self, other: Set[Any]) -> PeerAccessibleBySetProxy: ...
+    def __repr__(self) -> str: ...
     def _apply(self, additions, removals) -> None:
         """Compute the diff and issue a single ``cuMemPoolSetAccess``.
 
@@ -98,24 +69,12 @@ class PeerAccessibleBySetProxy:
         removals bypass that check (revoking is always permitted).
         """
 
-def replace_peer_accessible_by(mr: DeviceMemoryResource, devices: object) -> None:
-    """Replace the full peer-access set in a single batched driver call.
-
-    Backs the ``mr.peer_accessible_by = [...]`` setter. Uses the same planner
-    as the proxy's bulk ops; the only difference is that adds and removes are
-    derived from the symmetric difference between current driver state and the
-    requested target set.
-    """
-
 def normalize_peer_access_targets(owner_device_id: int, requested_devices: Iterable[object], *, resolve_device_id: Callable[[object], int]) -> tuple[int, ...]:
     """Return sorted, unique peer device IDs, excluding the owner device."""
-
 def plan_peer_access_update(owner_device_id: int, current_peer_ids: Iterable[int], requested_devices: Iterable[object], *, resolve_device_id: Callable[[object], int], can_access_peer: Callable[[int], bool]) -> PeerAccessPlan:
     """Compute the peer-access target state and add/remove deltas."""
-
 def _resolve_peer_device_id(value: Device | int | None) -> int:
     """Coerce ``Device | int`` into a device-ordinal int."""
-
 def _set_pool_access(mr: object, to_add: tuple[int, ...], to_remove: tuple[int, ...]) -> None:
     """Issue one ``cuMemPoolSetAccess`` for the given add/remove deltas.
 
@@ -127,7 +86,6 @@ def _set_pool_access(mr: object, to_add: tuple[int, ...], to_remove: tuple[int,
     Preconditions: ``len(to_add) + len(to_remove) > 0`` (the caller is
     responsible for skipping empty diffs).
     """
-
 def _apply_peer_access_diff(mr: DeviceMemoryResource, to_add: Iterable[int], to_remove: Iterable[int]) -> None:
     """Apply a peer-access diff in at most one driver call.
 
@@ -135,4 +93,12 @@ def _apply_peer_access_diff(mr: DeviceMemoryResource, to_add: Iterable[int], to_
     ``peer_accessible_by`` setter routes through this function. Empty diffs
     short-circuit here so the driver-level helper :func:`_set_pool_access` is
     only invoked when there is actual work for ``cuMemPoolSetAccess`` to do.
-    """
\ No newline at end of file
+    """
+def replace_peer_accessible_by(mr: DeviceMemoryResource, devices: object) -> None:
+    """Replace the full peer-access set in a single batched driver call.
+
+    Backs the ``mr.peer_accessible_by = [...]`` setter. Uses the same planner
+    as the proxy's bulk ops; the only difference is that adds and removes are
+    derived from the symmetric difference between current driver state and the
+    requested target set.
+    """
diff --git a/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyi b/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyi
index a74920a987f..76a7010dbc4 100644
--- a/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyi
+++ b/cuda_core/cuda/core/_memory/_pinned_memory_resource.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memory/_pinned_memory_resource.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memory/_pinned_memory_resource.pyx
 
 import uuid
 from dataclasses import dataclass
@@ -11,6 +9,7 @@ from cuda.core._memory._memory_pool import _MemPool
 from cuda.core._stream import Stream
 from cuda.core.graph import GraphBuilder
 
+__all__ = ['PinnedMemoryResource', 'PinnedMemoryResourceOptions']
 
 @dataclass
 class PinnedMemoryResourceOptions:
@@ -83,16 +82,10 @@ class PinnedMemoryResource(_MemPool):
 
     See :class:`DeviceMemoryResource` for more details on IPC usage patterns.
     """
-
-    def __init__(self, options: PinnedMemoryResourceOptions | None=None) -> None:
-        ...
-
+    def __init__(self, options: PinnedMemoryResourceOptions | None=None) -> None: ...
     def allocate(self, size: int, *, stream: Stream | GraphBuilder) -> Buffer:
         """Allocate a host-pinned buffer asynchronously on the supplied stream."""
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-
+    def __reduce__(self) -> tuple[object, ...]: ...
     @staticmethod
     def from_registry(uuid: uuid.UUID) -> PinnedMemoryResource:
         """
@@ -103,7 +96,6 @@ class PinnedMemoryResource(_MemPool):
         RuntimeError
             If no mapped memory resource is found in the registry.
         """
-
     def register(self, uuid: uuid.UUID) -> PinnedMemoryResource:
         """
         Register a mapped memory resource.
@@ -113,7 +105,6 @@ class PinnedMemoryResource(_MemPool):
         The registered mapped memory resource. If one was previously registered
         with the given key, it is returned.
         """
-
     @classmethod
     def from_allocation_handle(cls, alloc_handle: int | IPCAllocationHandle) -> PinnedMemoryResource:
         """Create a host-pinned memory resource from an allocation handle.
@@ -132,7 +123,6 @@ class PinnedMemoryResource(_MemPool):
         -------
             A new host-pinned memory resource instance with the imported handle.
         """
-
     @property
     def allocation_handle(self) -> IPCAllocationHandle:
         """Shareable handle for this memory pool (requires IPC).
@@ -140,23 +130,17 @@ class PinnedMemoryResource(_MemPool):
         The handle can be used to share the memory pool with other processes.
         The handle is cached in this `MemoryResource` and owned by it.
         """
-
     @property
     def device_id(self) -> int:
         """Return -1. Pinned memory is host memory and is not associated with a specific device."""
-
     @property
     def numa_id(self) -> int:
         """The host NUMA node ID used for pool placement, or -1 for OS-managed placement."""
-
     @property
     def is_device_accessible(self) -> bool:
         """Return True. This memory resource provides device-accessible buffers."""
-
     @property
     def is_host_accessible(self) -> bool:
         """Return True. This memory resource provides host-accessible buffers."""
-__all__ = ['PinnedMemoryResource', 'PinnedMemoryResourceOptions']
 
-def _deep_reduce_pinned_memory_resource(mr: object) -> tuple[object, ...]:
-    ...
\ No newline at end of file
+def _deep_reduce_pinned_memory_resource(mr: object) -> tuple[object, ...]: ...
diff --git a/cuda_core/cuda/core/_memory/_virtual_memory_resource.py b/cuda_core/cuda/core/_memory/_virtual_memory_resource.py
index 7cd12f597a6..4a0ec1f6bf7 100644
--- a/cuda_core/cuda/core/_memory/_virtual_memory_resource.py
+++ b/cuda_core/cuda/core/_memory/_virtual_memory_resource.py
@@ -350,10 +350,9 @@ def _grow_allocation_fast_path(
             # All succeeded, cancel undo actions
             trans.commit()
 
-        # Update the buffer size (pointer stays the same)
-        # TODO: #2049 This is a real bug, accessing _size which doesn't exist.
-        # Fix bug and remove the "type: ignore[attr-defined]" comment.
-        buf._size = new_size  # type: ignore[attr-defined]
+        # Update the buffer size (pointer stays the same). `Buffer.size` has
+        # no public setter, so this reaches into the private attribute.
+        buf._size = new_size
         return buf
 
     def _grow_allocation_slow_path(
diff --git a/cuda_core/cuda/core/_memoryview.pyi b/cuda_core/cuda/core/_memoryview.pyi
index e0ed0d3cf0d..e718be87140 100644
--- a/cuda_core/cuda/core/_memoryview.pyi
+++ b/cuda_core/cuda/core/_memoryview.pyi
@@ -1,10 +1,7 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_memoryview.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_memoryview.pyx
 
 import functools
-from collections.abc import Callable
-from typing import Any
+from typing import Any, Callable, TypedDict
 
 import numpy
 from cuda.core._layout import _StridedLayout
@@ -13,6 +10,10 @@ from cuda.core._stream import Stream
 
 from ._dlpack import *
 
+_SMV_DLPACK_EXCHANGE_API_CAPSULE = ...
+
+class PyTypeObject(TypedDict):
+    tp_dict: None
 
 class StridedMemoryView:
     """A class holding metadata of a strided dense array/tensor.
@@ -70,10 +71,13 @@ class StridedMemoryView:
         it will be the Buffer instance passed to the method.
 
     """
+    ptr: int
+    device_id: int
+    is_device_accessible: bool
+    readonly: bool
+    exporting_obj: object
 
-    def __init__(self, obj: object=None, stream_ptr: int | None=None) -> None:
-        ...
-
+    def __init__(self, obj: object | None=None, stream_ptr: int | None=None) -> None: ...
     @classmethod
     def from_dlpack(cls, obj: object, stream_ptr: int | None=None) -> StridedMemoryView:
         """Create a view from an object supporting the `DLPack <https://dmlc.github.io/dlpack/latest/>`_ protocol.
@@ -86,7 +90,6 @@ class StridedMemoryView:
         stream_ptr : int, optional
             Stream pointer for synchronization. If ``None``, no synchronization is performed.
         """
-
     @classmethod
     def from_cuda_array_interface(cls, obj: object, stream_ptr: int | None=None) -> StridedMemoryView:
         """Create a view from an object supporting the `__cuda_array_interface__ <https://numba.readthedocs.io/en/stable/cuda/cuda_array_interface.html>`_ protocol.
@@ -98,7 +101,6 @@ class StridedMemoryView:
         stream_ptr : int, optional
             Stream pointer for synchronization. If ``None``, no synchronization is performed.
         """
-
     @classmethod
     def from_array_interface(cls, obj: object) -> StridedMemoryView:
         """Create a view from an object supporting the `__array_interface__ <https://numpy.org/doc/stable/reference/arrays.interface.html>`_ protocol.
@@ -108,7 +110,6 @@ class StridedMemoryView:
         obj : object
             An object implementing the `__array_interface__ <https://numpy.org/doc/stable/reference/arrays.interface.html>`_ protocol (e.g., a numpy array).
         """
-
     @classmethod
     def from_any_interface(cls, obj: object, stream_ptr: int | None=None) -> StridedMemoryView:
         """Create a view by automatically selecting the best available protocol.
@@ -126,7 +127,6 @@ class StridedMemoryView:
         stream_ptr : int, optional
             Stream pointer for synchronization. If ``None``, no synchronization is performed.
         """
-
     @classmethod
     def from_buffer(cls, buffer: Buffer, shape: tuple[int, ...], strides: tuple[int, ...] | None=None, *, itemsize: int | None=None, dtype: numpy.dtype | None=None, is_readonly: bool=False) -> StridedMemoryView:
         """
@@ -155,17 +155,13 @@ class StridedMemoryView:
         is_readonly : bool, optional
             Whether the mark the view as readonly.
         """
-
-    def __dealloc__(self) -> None:
-        ...
-
+    def __dealloc__(self) -> None: ...
     def view(self, layout: _StridedLayout | None=None, dtype: numpy.dtype | None=None) -> StridedMemoryView:
         """
         Creates a new view with adjusted layout and dtype.
         Same as calling :meth:`from_buffer` with the current buffer.
         """
-
-    def as_tensor_map(self, box_dim: tuple[int, ...] | None=None, *, options: object=None, element_strides: tuple[int, ...] | None=None, data_type: object=None, interleave: object=None, swizzle: object=None, l2_promotion: object=None, oob_fill: object=None) -> object:
+    def as_tensor_map(self, box_dim: tuple[int, ...] | None=None, *, options: object | None=None, element_strides: tuple[int, ...] | None=None, data_type: object | None=None, interleave: object | None=None, swizzle: object | None=None, l2_promotion: object | None=None, oob_fill: object | None=None) -> object:
         """Create a tiled :obj:`TensorMapDescriptor` from this view.
 
         This is the public entry point for creating tiled tensor map
@@ -173,8 +169,7 @@ class StridedMemoryView:
         individual keyword arguments directly, or provide bundled tiled
         options via ``options=``.
         """
-
-    def copy_from(self, other: StridedMemoryView, stream: Stream, allocator: object=None, blocking: bool | None=None) -> None:
+    def copy_from(self, other: StridedMemoryView, stream: Stream, allocator: object | None=None, blocking: bool | None=None) -> None:
         """
         Copies the data from the other view into this view.
 
@@ -202,43 +197,32 @@ class StridedMemoryView:
                     * for device-to-device, it defaults to ``False`` (non-blocking),
                     * for host-to-device or device-to-host, it defaults to ``True`` (blocking).
         """
-
-    def copy_to(self, other: StridedMemoryView, stream: Stream | None=None, allocator: object=None, blocking: bool | None=None) -> None:
+    def copy_to(self, other: StridedMemoryView, stream: Stream | None=None, allocator: object | None=None, blocking: bool | None=None) -> None:
         """
         Copies the data from this view into the ``other`` view.
 
         For details, see :meth:`copy_from`.
         """
-
-    def __dlpack__(self, *, stream: int | None=None, max_version: tuple[int, int] | None=None, dl_device: tuple[int, int] | None=None, copy: bool | None=None) -> object:
-        ...
-
-    def __dlpack_device__(self) -> tuple[int, int]:
-        ...
-
+    def __dlpack__(self, *, stream: int | None=None, max_version: tuple[int, int] | None=None, dl_device: tuple[int, int] | None=None, copy: bool | None=None) -> object: ...
+    def __dlpack_device__(self) -> tuple[int, int]: ...
     @property
     def _layout(self) -> _StridedLayout:
         """
         The layout of the tensor. For StridedMemoryView created from DLPack or CAI,
         the layout is inferred from the tensor object's metadata.
         """
-
     @property
-    def size(self) -> int:
-        ...
-
+    def size(self) -> int: ...
     @property
     def shape(self) -> tuple[int, ...]:
         """
         Shape of the tensor.
         """
-
     @property
     def strides(self) -> tuple[int, ...] | None:
         """
         Strides of the tensor (in **counts**, not bytes).
         """
-
     @property
     def dtype(self) -> numpy.dtype | None:
         """
@@ -249,33 +233,21 @@ class StridedMemoryView:
         installed. If ``ml_dtypes`` is not available and such a tensor is encountered,
         a :obj:`NotImplementedError` will be raised.
         """
-
-    def __repr__(self) -> str:
-        ...
+    def __repr__(self) -> str: ...
 
 class _StridedMemoryViewProxy:
+    obj: object
+    has_dlpack: bool
 
-    def view(self, stream_ptr=None) -> StridedMemoryView:
-        ...
-
-    def __init__(self, obj: object) -> None:
-        ...
-_SMV_DLPACK_EXCHANGE_API_CAPSULE = ...
-
-def view_as_cai(obj, stream_ptr, view=None) -> StridedMemoryView:
-    ...
-
-def view_as_array_interface(obj, view=None) -> StridedMemoryView:
-    ...
+    def __init__(self, obj: object) -> None: ...
+    def view(self, stream_ptr=None) -> StridedMemoryView: ...
 
 @functools.lru_cache
-def _typestr2dtype(typestr: str) -> numpy.dtype:
-    ...
-
+def _typestr2dtype(typestr: str) -> numpy.dtype: ...
 @functools.lru_cache
-def _typestr2itemsize(typestr: str) -> int:
-    ...
-
+def _typestr2itemsize(typestr: str) -> int: ...
+def view_as_cai(obj, stream_ptr, view=None) -> StridedMemoryView: ...
+def view_as_array_interface(obj, view=None) -> StridedMemoryView: ...
 def args_viewable_as_strided_memory(arg_indices: tuple[int, ...]) -> Callable[[Callable[..., Any]], Callable[..., Any]]:
     """
     Decorator to create proxy objects to :obj:`StridedMemoryView` for the
@@ -304,4 +276,4 @@ def args_viewable_as_strided_memory(arg_indices: tuple[int, ...]) -> Callable[[C
     ----------
     arg_indices : tuple
         The indices of the target positional arguments.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/_module.pyi b/cuda_core/cuda/core/_module.pyi
index e8604af012c..9fb502920ad 100644
--- a/cuda_core/cuda/core/_module.pyi
+++ b/cuda_core/cuda/core/_module.pyi
@@ -1,16 +1,15 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_module.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_module.pyx
 
-from __future__ import annotations
-
-from collections import namedtuple
 from os import PathLike
+from typing import Any
 
-import cython
 from cuda.core._device import Device
 from cuda.core._launch_config import LaunchConfig
 from cuda.core._stream import Stream
 from cuda.core._utils.cuda_utils import driver
 
+__all__ = ['Kernel', 'ObjectCode']
+CodeTypeT = bytes | bytearray | str
 
 class KernelAttributes:
     """Read-only view of a kernel's per-device attributes.
@@ -22,10 +21,7 @@ class KernelAttributes:
     views share the underlying cache so a value queried through one view
     is visible through the others.
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     def __getitem__(self, device: Device | int) -> KernelAttributes:
         """Return a view of these attributes bound to a specific device.
 
@@ -41,77 +37,61 @@ class KernelAttributes:
             A view bound to ``device`` that shares the underlying cache
             with this view.
         """
-
     @property
     def max_threads_per_block(self) -> int:
         """int : The maximum number of threads per block.
         This attribute is read-only."""
-
     @property
     def shared_size_bytes(self) -> int:
         """int : The size in bytes of statically-allocated shared memory required by this function.
         This attribute is read-only."""
-
     @property
     def const_size_bytes(self) -> int:
         """int : The size in bytes of user-allocated constant memory required by this function.
         This attribute is read-only."""
-
     @property
     def local_size_bytes(self) -> int:
         """int : The size in bytes of local memory used by each thread of this function.
         This attribute is read-only."""
-
     @property
     def num_regs(self) -> int:
         """int : The number of registers used by each thread of this function.
         This attribute is read-only."""
-
     @property
     def ptx_version(self) -> int:
         """int : The PTX virtual architecture version for which the function was compiled.
         This attribute is read-only."""
-
     @property
     def binary_version(self) -> int:
         """int : The binary architecture version for which the function was compiled.
         This attribute is read-only."""
-
     @property
     def cache_mode_ca(self) -> bool:
         """bool : Whether the function has been compiled with user specified option "-Xptxas --dlcm=ca" set.
         This attribute is read-only."""
-
     @property
     def max_dynamic_shared_size_bytes(self) -> int:
         """int : The maximum size in bytes of dynamically-allocated shared memory that can be used
         by this function."""
-
     @property
     def preferred_shared_memory_carveout(self) -> int:
         """int : The shared memory carveout preference, in percent of the total shared memory."""
-
     @property
     def cluster_size_must_be_set(self) -> bool:
         """bool : The kernel must launch with a valid cluster size specified.
         This attribute is read-only."""
-
     @property
     def required_cluster_width(self) -> int:
         """int : The required cluster width in blocks."""
-
     @property
     def required_cluster_height(self) -> int:
         """int : The required cluster height in blocks."""
-
     @property
     def required_cluster_depth(self) -> int:
         """int : The required cluster depth in blocks."""
-
     @property
     def non_portable_cluster_size_allowed(self) -> bool:
         """bool : Whether the function can be launched with non-portable cluster size."""
-
     @property
     def cluster_scheduling_policy_preference(self) -> int:
         """int : The block scheduling policy of a function."""
@@ -120,10 +100,7 @@ class KernelOccupancy:
     """This class offers methods to query occupancy metrics that help determine optimal
     launch parameters such as block size, grid size, and shared memory usage.
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     def max_active_blocks_per_multiprocessor(self, block_size: int, dynamic_shared_memory_size: int) -> int:
         """Occupancy of the kernel.
 
@@ -149,8 +126,7 @@ class KernelOccupancy:
             theoretical multiprocessor utilization (occupancy).
 
         """
-
-    def max_potential_block_size(self, dynamic_shared_memory_needed: int | driver.CUoccupancyB2DSize, block_size_limit: int) -> MaxPotentialBlockSizeOccupancyResult:
+    def max_potential_block_size(self, dynamic_shared_memory_needed: int | driver.CUoccupancyB2DSize, block_size_limit: int) -> Any:
         """MaxPotentialBlockSizeOccupancyResult: Suggested launch configuration for reasonable occupancy.
 
         Returns the minimum grid size needed to achieve the maximum occupancy and
@@ -180,7 +156,6 @@ class KernelOccupancy:
             Interpreter Lock may lead to deadlocks.
 
         """
-
     def available_dynamic_shared_memory_per_block(self, num_blocks_per_multiprocessor: int, block_size: int) -> int:
         """Dynamic shared memory available per block for given launch configuration.
 
@@ -198,7 +173,6 @@ class KernelOccupancy:
         int
             Dynamic shared memory available per block for given launch configuration.
         """
-
     def max_potential_cluster_size(self, config: LaunchConfig, *, stream: Stream) -> int:
         """Maximum potential cluster size.
 
@@ -218,7 +192,6 @@ class KernelOccupancy:
         int
             The maximum cluster size that can be launched for this kernel and launch configuration.
         """
-
     def max_active_clusters(self, config: LaunchConfig, *, stream: Stream) -> int:
         """Maximum number of active clusters on the target device.
 
@@ -249,28 +222,19 @@ class Kernel:
     should instead be created through a :obj:`~_module.ObjectCode` object.
 
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @property
-    @cython.critical_section
     def attributes(self) -> KernelAttributes:
         """Get the read-only attributes of this kernel."""
-
     @property
     def num_arguments(self) -> int:
         """int : The number of arguments of this function"""
-
     @property
-    def arguments_info(self) -> list[ParamInfo]:
+    def arguments_info(self) -> list[Any]:
         """list[ParamInfo]: (offset, size) for each argument of this function"""
-
     @property
-    @cython.critical_section
     def occupancy(self) -> KernelOccupancy:
         """Get the occupancy information for launching this kernel."""
-
     @property
     def handle(self) -> object:
         """Return the underlying kernel handle object.
@@ -280,11 +244,8 @@ class Kernel:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(Kernel.handle)``.
         """
-
     @property
-    def _handle(self) -> object:
-        ...
-
+    def _handle(self) -> object: ...
     @staticmethod
     def from_handle(handle, mod: ObjectCode | None=None) -> Kernel:
         """Creates a new :obj:`Kernel` object from a kernel handle.
@@ -299,15 +260,9 @@ class Kernel:
             library lifetime for foreign kernels not created by
             cuda.core.
         """
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
-    def __repr__(self) -> str:
-        ...
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
+    def __repr__(self) -> str: ...
 
 class ObjectCode:
     """Represent a compiled program to be loaded onto the device.
@@ -322,21 +277,12 @@ class ObjectCode:
     from all other possible code types should be avoided in favor of compilation through
     :class:`~cuda.core.Program`
     """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @classmethod
-    def _init(cls, module, code_type, *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
-        ...
-
+    def _init(cls, module, code_type, *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode: ...
     @staticmethod
-    def _reduce_helper(module, code_type, name, symbol_mapping):
-        ...
-
-    def __reduce__(self) -> tuple[object, ...]:
-        ...
-
+    def _reduce_helper(module, code_type, name, symbol_mapping): ...
+    def __reduce__(self) -> tuple[object, ...]: ...
     @staticmethod
     def from_cubin(module: bytes | str | PathLike[str], *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
         """Create an :class:`ObjectCode` instance from an existing cubin.
@@ -354,7 +300,6 @@ class ObjectCode:
             should be mapped to the mangled names before trying to retrieve
             them (default to no mappings).
         """
-
     @staticmethod
     def from_ptx(module: bytes | str | PathLike[str], *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
         """Create an :class:`ObjectCode` instance from an existing PTX.
@@ -372,7 +317,6 @@ class ObjectCode:
             should be mapped to the mangled names before trying to retrieve
             them (default to no mappings).
         """
-
     @staticmethod
     def from_ltoir(module: bytes | str | PathLike[str], *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
         """Create an :class:`ObjectCode` instance from an existing LTOIR.
@@ -390,7 +334,6 @@ class ObjectCode:
             should be mapped to the mangled names before trying to retrieve
             them (default to no mappings).
         """
-
     @staticmethod
     def from_fatbin(module: bytes | str | PathLike[str], *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
         """Create an :class:`ObjectCode` instance from an existing fatbin.
@@ -408,7 +351,6 @@ class ObjectCode:
             should be mapped to the mangled names before trying to retrieve
             them (default to no mappings).
         """
-
     @staticmethod
     def from_object(module: bytes | str, *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
         """Create an :class:`ObjectCode` instance from an existing object code.
@@ -425,7 +367,6 @@ class ObjectCode:
             should be mapped to the mangled names before trying to retrieve
             them (default to no mappings).
         """
-
     @staticmethod
     def from_library(module: bytes | str, *, name: str='', symbol_mapping: dict[str, str] | None=None) -> ObjectCode:
         """Create an :class:`ObjectCode` instance from an existing library.
@@ -442,7 +383,6 @@ class ObjectCode:
             should be mapped to the mangled names before trying to retrieve
             them (default to no mappings).
         """
-
     def get_kernel(self, name: str | bytes) -> Kernel:
         """Return the :obj:`~_module.Kernel` of a specified name from this object code.
 
@@ -457,7 +397,6 @@ class ObjectCode:
             Newly created kernel object.
 
         """
-
     def get_module(self) -> driver.CUmodule:
         """Return a context-dependent :obj:`~driver.CUmodule` for legacy interop.
 
@@ -471,23 +410,18 @@ class ObjectCode:
             Module handle for the current CUDA context, suitable for legacy
             driver APIs that accept ``CUmodule``.
         """
-
     @property
     def code(self) -> CodeTypeT:
         """Return the underlying code object."""
-
     @property
     def name(self) -> str:
         """Return a human-readable name of this code object."""
-
     @property
     def code_type(self) -> str:
         """Return the type of the underlying code object."""
-
     @property
     def symbol_mapping(self) -> dict[str, str]:
         """Return a copy of the symbol mapping dictionary."""
-
     @property
     def handle(self) -> object:
         """Return the native, context-independent :obj:`~driver.CUlibrary` handle.
@@ -500,16 +434,6 @@ class ObjectCode:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(ObjectCode.handle)``.
         """
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-__all__ = ['Kernel', 'ObjectCode']
-MaxPotentialBlockSizeOccupancyResult = namedtuple('MaxPotentialBlockSizeOccupancyResult', ('min_grid_size', 'max_block_size'))
-ParamInfo = namedtuple('ParamInfo', ['offset', 'size'])
-CodeTypeT = bytes | bytearray | str
\ No newline at end of file
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
+    def __repr__(self) -> str: ...
diff --git a/cuda_core/cuda/core/_program.pyi b/cuda_core/cuda/core/_program.pyi
index fd40aae069f..3e525597977 100644
--- a/cuda_core/cuda/core/_program.pyi
+++ b/cuda_core/cuda/core/_program.pyi
@@ -1,12 +1,10 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_program.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_program.pyx
 
 """Compilation machinery for CUDA programs.
 
 This module provides :class:`Program` for compiling source code into
 :class:`~cuda.core.ObjectCode`, with :class:`ProgramOptions` for configuration.
 """
-from __future__ import annotations
-
 from dataclasses import dataclass
 
 from cuda.bindings import nvrtc
@@ -16,6 +14,10 @@ from cuda.core.typing import (CompilerBackendType, ObjectCodeFormatType,
                               PCHStatusType, SourceCodeType)
 from cuda.core.utils._program_cache import ProgramCacheResource
 
+__all__ = ['Program', 'ProgramOptions']
+ProgramHandleT = nvrtc.nvrtcProgram | int | LinkerHandleT
+_nvvm_module = None
+_nvvm_import_attempted = False
 
 class Program:
     """Represent a compilation machinery to process programs into
@@ -35,14 +37,10 @@ class Program:
     options : :class:`ProgramOptions`, optional
         Options to customize the compilation process.
     """
-
-    def __init__(self, code: str | bytes | bytearray, code_type: SourceCodeType | str, options: ProgramOptions | None=None):
-        ...
-
+    def __init__(self, code: str | bytes | bytearray, code_type: SourceCodeType | str, options: ProgramOptions | None=None): ...
     def close(self) -> None:
         """Destroy this program."""
-
-    def compile(self, target_type: ObjectCodeFormatType | str, name_expressions: tuple[str, ...] | list[str]=..., logs: object=None, *, cache: ProgramCacheResource | None=None) -> ObjectCode:
+    def compile(self, target_type: ObjectCodeFormatType | str, name_expressions: tuple[str, ...] | list[str]=(), logs: object | None=None, *, cache: ProgramCacheResource | None=None) -> ObjectCode:
         """Compile the program to the specified target type.
 
         Parameters
@@ -90,7 +88,6 @@ class Program:
         :class:`~cuda.core.ObjectCode`
             The compiled object code.
         """
-
     @property
     def pch_status(self) -> PCHStatusType | None:
         """PCH creation outcome from the most recent :meth:`compile` call.
@@ -115,11 +112,9 @@ class Program:
            use the NVRTC backend. For PTX and NVVM programs this property
            always returns ``None``.
         """
-
     @property
     def backend(self) -> CompilerBackendType:
         """Return this Program instance's underlying :class:`CompilerBackendType`."""
-
     @property
     def handle(self) -> ProgramHandleT:
         """Return the underlying handle object.
@@ -133,9 +128,7 @@ class Program:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(Program.handle)``.
         """
-
-    def __repr__(self) -> str:
-        ...
+    def __repr__(self) -> str: ...
 
 @dataclass
 class ProgramOptions:
@@ -374,15 +367,9 @@ class ProgramOptions:
     use_libdevice: bool | None = None
     numba_debug: bool | None = None
 
-    def __post_init__(self) -> None:
-        ...
-
-    def _prepare_nvrtc_options(self) -> list[bytes]:
-        ...
-
-    def _prepare_nvvm_options(self, as_bytes: bool=True) -> list[bytes] | list[str]:
-        ...
-
+    def __post_init__(self) -> None: ...
+    def _prepare_nvrtc_options(self) -> list[bytes]: ...
+    def _prepare_nvvm_options(self, as_bytes: bool=True) -> list[bytes] | list[str]: ...
     def as_bytes(self, backend: CompilerBackendType | str, target_type: ObjectCodeFormatType | str | None=None) -> list[bytes]:
         """Convert program options to bytes format for the specified backend.
 
@@ -415,30 +402,9 @@ class ProgramOptions:
         >>> options = ProgramOptions(arch="sm_80", debug=True)
         >>> nvrtc_options = options.as_bytes("nvrtc")
         """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def _prepare_extra_sources_bytes(self) -> list[tuple[bytes, bytes]] | None:
         """Convert extra_sources to bytes format for NVVM."""
-__all__ = ['Program', 'ProgramOptions']
-ProgramHandleT = nvrtc.nvrtcProgram | int | LinkerHandleT
-_nvvm_module = None
-_nvvm_import_attempted = False
-
-def _can_load_generated_ptx() -> bool:
-    """Check if the driver can load PTX generated by the current NVRTC version."""
-
-def _assert_single_dashed_nvvm_options(options: list[str]) -> None:
-    """Guard against emitting a double-dashed option to libNVVM.
-
-    libNVVM's parser accepts only single-dashed options and rejects the
-    double-dashed spelling of every option with NVVM_ERROR_INVALID_OPTION
-    (see #2570). Every option on this path is generated from typed fields, so
-    a double dash can only mean a bug in ``cuda.core`` rather than bad user
-    input. Fail here, naming the option, instead of leaving the user with
-    libNVVM's opaque error.
-    """
 
 def _program_compile_uncached(program, target_type, name_expressions, logs):
     """Run ``Program_compile`` without the cache wrapper.
@@ -449,9 +415,19 @@ def _program_compile_uncached(program, target_type, name_expressions, logs):
     and its methods cannot be reassigned from Python, so the seam must live
     outside the class.
     """
-
 def _get_nvvm_module() -> object:
     """Get the NVVM module, importing it lazily with availability checks."""
-
 def _find_libdevice_path() -> object:
-    """Find libdevice*.bc for NVVM compilation using cuda.pathfinder."""
\ No newline at end of file
+    """Find libdevice*.bc for NVVM compilation using cuda.pathfinder."""
+def _can_load_generated_ptx() -> bool:
+    """Check if the driver can load PTX generated by the current NVRTC version."""
+def _assert_single_dashed_nvvm_options(options: list[str]) -> None:
+    """Guard against emitting a double-dashed option to libNVVM.
+
+    libNVVM's parser accepts only single-dashed options and rejects the
+    double-dashed spelling of every option with NVVM_ERROR_INVALID_OPTION
+    (see #2570). Every option on this path is generated from typed fields, so
+    a double dash can only mean a bug in ``cuda.core`` rather than bad user
+    input. Fail here, naming the option, instead of leaving the user with
+    libNVVM's opaque error.
+    """
diff --git a/cuda_core/cuda/core/_resource_handles.pyi b/cuda_core/cuda/core/_resource_handles.pyi
index f9b10d4db3d..457e2921047 100644
--- a/cuda_core/cuda/core/_resource_handles.pyi
+++ b/cuda_core/cuda/core/_resource_handles.pyi
@@ -1,31 +1,29 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_resource_handles.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_resource_handles.pyx
 
-from __future__ import annotations
+from typing import Any, TypeAlias
 
-from libcpp.memory import shared_ptr, unique_ptr
-
-ContextHandle = shared_ptr
-GreenCtxHandle = shared_ptr
-StreamHandle = shared_ptr
-EventHandle = shared_ptr
-MemoryPoolHandle = shared_ptr
-DevicePtrHandle = shared_ptr
-LibraryHandle = shared_ptr
-KernelHandle = shared_ptr
-GraphHandle = shared_ptr
-GraphExecHandle = shared_ptr
-GraphNodeHandle = shared_ptr
-GraphicsResourceHandle = shared_ptr
-NvrtcProgramHandle = shared_ptr
-NvvmProgramHandle = shared_ptr
-NvJitLinkHandle = shared_ptr
-CuLinkHandle = shared_ptr
-FileDescriptorHandle = shared_ptr
-OpaqueArrayHandle = shared_ptr
-MipmappedArrayHandle = shared_ptr
-TexObjectHandle = shared_ptr
-SurfObjectHandle = shared_ptr
-OpaqueHandle = shared_ptr
-PreparedAttachment = unique_ptr
-PreparedChildGraphUpdate = shared_ptr
-PreparedExecAttachment = unique_ptr
\ No newline at end of file
+ContextHandle: TypeAlias = Any
+GreenCtxHandle: TypeAlias = Any
+StreamHandle: TypeAlias = Any
+EventHandle: TypeAlias = Any
+MemoryPoolHandle: TypeAlias = Any
+DevicePtrHandle: TypeAlias = Any
+LibraryHandle: TypeAlias = Any
+KernelHandle: TypeAlias = Any
+GraphHandle: TypeAlias = Any
+GraphExecHandle: TypeAlias = Any
+GraphNodeHandle: TypeAlias = Any
+GraphicsResourceHandle: TypeAlias = Any
+NvrtcProgramHandle: TypeAlias = Any
+NvvmProgramHandle: TypeAlias = Any
+NvJitLinkHandle: TypeAlias = Any
+CuLinkHandle: TypeAlias = Any
+FileDescriptorHandle: TypeAlias = Any
+OpaqueArrayHandle: TypeAlias = Any
+MipmappedArrayHandle: TypeAlias = Any
+TexObjectHandle: TypeAlias = Any
+SurfObjectHandle: TypeAlias = Any
+OpaqueHandle: TypeAlias = Any
+PreparedAttachment: TypeAlias = Any
+PreparedChildGraphUpdate: TypeAlias = Any
+PreparedExecAttachment: TypeAlias = Any
diff --git a/cuda_core/cuda/core/_stream.pyi b/cuda_core/cuda/core/_stream.pyi
index 99af5f9b15b..bee6efd9d31 100644
--- a/cuda_core/cuda/core/_stream.pyi
+++ b/cuda_core/cuda/core/_stream.pyi
@@ -1,18 +1,18 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_stream.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_stream.pyx
 
 from dataclasses import dataclass
-from typing import Protocol
+from typing import Any, Protocol
 
 import cuda.bindings.driver
-import cython
 from cuda.core._context import Context
 from cuda.core._device import Device
 from cuda.core._device_resources import DeviceResources
 from cuda.core._event import Event, EventOptions
 from cuda.core.graph import GraphBuilder
 
+__all__ = ['LEGACY_DEFAULT_STREAM', 'PER_THREAD_DEFAULT_STREAM', 'Stream', 'StreamOptions']
+LEGACY_DEFAULT_STREAM: Stream = Stream._legacy_default()
+PER_THREAD_DEFAULT_STREAM: Stream = Stream._per_thread_default()
 
 @dataclass
 class StreamOptions:
@@ -27,11 +27,10 @@ class StreamOptions:
         higher priority. (Default to lowest priority)
 
     """
-    nonblocking: cython.bint = True
+    nonblocking: Any = True
     priority: int | None = None
 
 class IsStreamType(Protocol):
-
     def __cuda_stream__(self) -> tuple[int, int]:
         """
         For any Python object that is meant to be interpreted as a CUDA stream, the intent
@@ -56,42 +55,27 @@ class Stream:
     object, or created directly through using an existing handle
     using Stream.from_handle().
     """
-
-    def close(self):
-        """Destroy the stream.
-
-        Releases the stream handle. For owned streams, this destroys the
-        underlying CUDA stream. For borrowed streams, this releases the
-        reference and allows the Python owner to be GC'd.
-        """
-
-    def __init__(self, *args, **kwargs) -> None:
-        ...
-
+    def __init__(self, *args, **kwargs) -> None: ...
     @classmethod
     def _legacy_default(cls) -> Stream:
         """Return the legacy default stream (supports subclassing)."""
-
     @classmethod
     def _per_thread_default(cls) -> Stream:
         """Return the per-thread default stream (supports subclassing)."""
-
     @classmethod
-    def _init(cls, obj: IsStreamType | None=None, options: object=None, device_id: int | None=None, ctx: Context | None=None) -> Stream:
-        ...
+    def _init(cls, obj: IsStreamType | None=None, options: object | None=None, device_id: int | None=None, ctx: Context | None=None) -> Stream: ...
+    def close(self):
+        """Destroy the stream.
 
+        Releases the stream handle. For owned streams, this destroys the
+        underlying CUDA stream. For borrowed streams, this releases the
+        reference and allows the Python owner to be GC'd.
+        """
     def __cuda_stream__(self) -> tuple[int, int]:
         """Return an instance of a __cuda_stream__ protocol."""
-
-    def __hash__(self) -> int:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __repr__(self) -> str:
-        ...
-
+    def __hash__(self) -> int: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __repr__(self) -> str: ...
     @property
     def handle(self) -> cuda.bindings.driver.CUstream:
         """Return the underlying ``CUstream`` object.
@@ -101,18 +85,14 @@ class Stream:
             This handle is a Python object. To get the memory address of the underlying C
             handle, call ``int(Stream.handle)``.
         """
-
     @property
     def is_nonblocking(self) -> bool:
         """Return True if this is a nonblocking stream, otherwise False."""
-
     @property
     def priority(self) -> int:
         """Return the stream priority."""
-
     def sync(self) -> None:
         """Synchronize the stream."""
-
     def record(self, event: Event | None=None, options: EventOptions | None=None) -> Event:
         """Record an event onto the stream.
 
@@ -139,7 +119,6 @@ class Stream:
         associated with the current context at call time.
 
         """
-
     def wait(self, event_or_stream: Event | Stream) -> None:
         """Wait for a CUDA event or a CUDA stream.
 
@@ -157,7 +136,6 @@ class Stream:
             streams.
 
         """
-
     @property
     def device(self) -> Device:
         """Return the :obj:`~_device.Device` singleton associated with this stream.
@@ -174,7 +152,6 @@ class Stream:
         set current after the stream is created.
 
         """
-
     @property
     def context(self) -> Context:
         """Return the :obj:`~_context.Context` associated with this stream.
@@ -187,7 +164,6 @@ class Stream:
         context at call time.
 
         """
-
     @property
     def resources(self) -> DeviceResources:
         """Query the hardware resources provisioned for this stream's context.
@@ -204,7 +180,6 @@ class Stream:
         context at call time.
 
         """
-
     @staticmethod
     def from_handle(handle) -> Stream:
         """Create a new :obj:`~_stream.Stream` object from a foreign stream handle.
@@ -229,7 +204,6 @@ class Stream:
             Newly created stream object.
 
         """
-
     def create_graph_builder(self) -> GraphBuilder:
         """Create a new :obj:`~graph.GraphBuilder` object.
 
@@ -241,9 +215,6 @@ class Stream:
             Newly created graph builder object.
 
         """
-__all__ = ['LEGACY_DEFAULT_STREAM', 'PER_THREAD_DEFAULT_STREAM', 'Stream', 'StreamOptions']
-LEGACY_DEFAULT_STREAM: Stream = Stream._legacy_default()
-PER_THREAD_DEFAULT_STREAM: Stream = Stream._per_thread_default()
 
 def default_stream() -> Stream:
     """Return the default CUDA :obj:`~_stream.Stream`.
@@ -255,6 +226,4 @@ def default_stream() -> Stream:
     the legacy stream.
 
     """
-
-def Stream_accept(arg, allow_stream_protocol: bool=False) -> Stream:
-    ...
\ No newline at end of file
+def Stream_accept(arg, allow_stream_protocol: bool=False) -> Stream: ...
diff --git a/cuda_core/cuda/core/_tensor_bridge.pyi b/cuda_core/cuda/core/_tensor_bridge.pyi
index 22948d5b864..25c13c6458c 100644
--- a/cuda_core/cuda/core/_tensor_bridge.pyi
+++ b/cuda_core/cuda/core/_tensor_bridge.pyi
@@ -1,4 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_tensor_bridge.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_tensor_bridge.pyx
 
 """Tensor bridge: extract PyTorch tensor metadata via the AOTI stable C ABI.
 
@@ -46,13 +46,19 @@ Credit: Emilio Castillo (ecastillo@nvidia.com) – original tensor-bridge POC.
    ``torch._C`` has been re-opened with ``RTLD_GLOBAL`` *before*
    importing this module so that the AOTI symbols are visible.
 """
-from __future__ import annotations
+from typing import TypeAlias, TypedDict
 
 import numpy
 from cuda.core._memoryview import StridedMemoryView
 
-AOTITorchError = int
+AOTITorchError: TypeAlias = int
 
+class PyObject(TypedDict): ...
+
+class AtenTensorOpaque(TypedDict): ...
+
+def resolve_aoti_dtype(dtype_code: int) -> numpy.dtype:
+    """Python-callable wrapper around _get_aoti_dtype (for lazy resolution)."""
 def sync_torch_stream(device_index: int, consumer_s: int) -> int:
     """Establish stream ordering between PyTorch's current CUDA stream
     and the given consumer stream.
@@ -61,10 +67,6 @@ def sync_torch_stream(device_index: int, consumer_s: int) -> int:
     the consumer stream wait on it.  This is a no-op if both streams are
     the same.
     """
-
-def resolve_aoti_dtype(dtype_code: int) -> numpy.dtype:
-    """Python-callable wrapper around _get_aoti_dtype (for lazy resolution)."""
-
 def view_as_torch_tensor(obj: object, stream_ptr: object, view: StridedMemoryView | None=None) -> StridedMemoryView:
     """Create/populate a :class:`StridedMemoryView` from a ``torch.Tensor``.
 
@@ -82,4 +84,4 @@ def view_as_torch_tensor(obj: object, stream_ptr: object, view: StridedMemoryVie
     view : StridedMemoryView, optional
         If provided, populate this existing view in-place.  Otherwise a
         new instance is created.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/_tensor_map.pyi b/cuda_core/cuda/core/_tensor_map.pyi
index c6a18ad2399..2a3dc8a48a3 100644
--- a/cuda_core/cuda/core/_tensor_map.pyi
+++ b/cuda_core/cuda/core/_tensor_map.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_tensor_map.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_tensor_map.pyx
 
 from dataclasses import dataclass
 
@@ -8,6 +6,22 @@ import numpy
 from cuda.bindings import cydriver
 from cuda.core._device import Device
 
+__all__ = ['TensorMapDescriptor', 'TensorMapDescriptorOptions']
+_TMA_DT_UINT8: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT8)
+_TMA_DT_UINT16: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT16)
+_TMA_DT_UINT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT32)
+_TMA_DT_INT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT32)
+_TMA_DT_UINT64: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT64)
+_TMA_DT_INT64: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT64)
+_TMA_DT_FLOAT16: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT16)
+_TMA_DT_FLOAT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32)
+_TMA_DT_FLOAT64: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT64)
+_TMA_DT_BFLOAT16: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_BFLOAT16)
+_TMA_DT_FLOAT32_FTZ: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32_FTZ)
+_TMA_DT_TFLOAT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32)
+_TMA_DT_TFLOAT32_FTZ: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32_FTZ)
+_NUMPY_DTYPE_TO_TMA = {numpy.dtype(numpy.uint8): _TMA_DT_UINT8, numpy.dtype(numpy.uint16): _TMA_DT_UINT16, numpy.dtype(numpy.uint32): _TMA_DT_UINT32, numpy.dtype(numpy.int32): _TMA_DT_INT32, numpy.dtype(numpy.uint64): _TMA_DT_UINT64, numpy.dtype(numpy.int64): _TMA_DT_INT64, numpy.dtype(numpy.float16): _TMA_DT_FLOAT16, numpy.dtype(numpy.float32): _TMA_DT_FLOAT32, numpy.dtype(numpy.float64): _TMA_DT_FLOAT64}
+_TMA_DATA_TYPE_SIZE = {_TMA_DT_UINT8: 1, _TMA_DT_UINT16: 2, _TMA_DT_UINT32: 4, _TMA_DT_INT32: 4, _TMA_DT_UINT64: 8, _TMA_DT_INT64: 8, _TMA_DT_FLOAT16: 2, _TMA_DT_FLOAT32: 4, _TMA_DT_FLOAT64: 8, _TMA_DT_BFLOAT16: 2, _TMA_DT_FLOAT32_FTZ: 4, _TMA_DT_TFLOAT32: 4, _TMA_DT_TFLOAT32_FTZ: 4}
 
 class TensorMapDataType:
     """Data types for tensor map descriptors.
@@ -105,8 +119,7 @@ class TensorMapDescriptorOptions:
     l2_promotion: TensorMapL2Promotion = TensorMapL2Promotion.NONE
     oob_fill: TensorMapOOBFill = TensorMapOOBFill.NONE
 
-    def __post_init__(self) -> None:
-        ...
+    def __post_init__(self) -> None: ...
 
 class TensorMapDescriptor:
     """Describes a TMA (Tensor Memory Accelerator) tensor map for Hopper+ GPUs.
@@ -121,16 +134,12 @@ class TensorMapDescriptor:
     descriptors can be passed directly to :func:`~cuda.core.launch` as a
     kernel argument.
     """
-
-    def __init__(self):
-        ...
-
+    def __init__(self): ...
     @property
     def device(self) -> Device | None:
         """Return the :obj:`~cuda.core.Device` associated with this descriptor."""
-
     @classmethod
-    def _from_tiled(cls, view, box_dim=None, *, options=None, element_strides=None, data_type=None, interleave=..., swizzle=..., l2_promotion=..., oob_fill=...):
+    def _from_tiled(cls, view, box_dim=None, *, options=None, element_strides=None, data_type=None, interleave=TensorMapInterleave.NONE, swizzle=TensorMapSwizzle.NONE, l2_promotion=TensorMapL2Promotion.NONE, oob_fill=TensorMapOOBFill.NONE):
         """Create a tiled TMA descriptor from a validated view.
 
         Parameters
@@ -171,9 +180,8 @@ class TensorMapDescriptor:
             If the tensor rank is outside [1, 5], the pointer is not
             16-byte aligned, or dimension/stride constraints are violated.
         """
-
     @classmethod
-    def _from_im2col(cls, view, pixel_box_lower_corner, pixel_box_upper_corner, channels_per_pixel, pixels_per_column, *, element_strides=None, data_type=None, interleave=..., swizzle=..., l2_promotion=..., oob_fill=...):
+    def _from_im2col(cls, view, pixel_box_lower_corner, pixel_box_upper_corner, channels_per_pixel, pixels_per_column, *, element_strides=None, data_type=None, interleave=TensorMapInterleave.NONE, swizzle=TensorMapSwizzle.NONE, l2_promotion=TensorMapL2Promotion.NONE, oob_fill=TensorMapOOBFill.NONE):
         """Create an im2col TMA descriptor from a validated view.
 
         Im2col layout is used for convolution-style data access patterns.
@@ -219,9 +227,8 @@ class TensorMapDescriptor:
             If the tensor rank is outside [3, 5], the pointer is not
             16-byte aligned, or other constraints are violated.
         """
-
     @classmethod
-    def _from_im2col_wide(cls, view, pixel_box_lower_corner_width, pixel_box_upper_corner_width, channels_per_pixel, pixels_per_column, *, element_strides=None, data_type=None, interleave=..., mode=..., swizzle=..., l2_promotion=..., oob_fill=...):
+    def _from_im2col_wide(cls, view, pixel_box_lower_corner_width, pixel_box_upper_corner_width, channels_per_pixel, pixels_per_column, *, element_strides=None, data_type=None, interleave=TensorMapInterleave.NONE, mode=TensorMapIm2ColWideMode.W, swizzle=TensorMapSwizzle.SWIZZLE_128B, l2_promotion=TensorMapL2Promotion.NONE, oob_fill=TensorMapOOBFill.NONE):
         """Create an im2col-wide TMA descriptor from a validated view.
 
         Im2col-wide layout loads elements exclusively along the W (width)
@@ -267,7 +274,6 @@ class TensorMapDescriptor:
             If the tensor rank is outside [3, 5], the pointer is not
             16-byte aligned, or other constraints are violated.
         """
-
     def replace_address(self, tensor: object) -> None:
         """Replace the global memory address in this tensor map descriptor.
 
@@ -281,44 +287,16 @@ class TensorMapDescriptor:
             or a :obj:`~cuda.core.StridedMemoryView`. Must refer to
             device-accessible memory with a 16-byte-aligned pointer.
         """
+    def __repr__(self) -> str: ...
 
-    def __repr__(self) -> str:
-        ...
-__all__ = ['TensorMapDescriptor', 'TensorMapDescriptorOptions']
-_TMA_DT_UINT8 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT8)
-_TMA_DT_UINT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT16)
-_TMA_DT_UINT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT32)
-_TMA_DT_INT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT32)
-_TMA_DT_UINT64 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT64)
-_TMA_DT_INT64 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT64)
-_TMA_DT_FLOAT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT16)
-_TMA_DT_FLOAT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32)
-_TMA_DT_FLOAT64 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT64)
-_TMA_DT_BFLOAT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_BFLOAT16)
-_TMA_DT_FLOAT32_FTZ = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32_FTZ)
-_TMA_DT_TFLOAT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32)
-_TMA_DT_TFLOAT32_FTZ = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32_FTZ)
-_NUMPY_DTYPE_TO_TMA = {numpy.dtype(numpy.uint8): _TMA_DT_UINT8, numpy.dtype(numpy.uint16): _TMA_DT_UINT16, numpy.dtype(numpy.uint32): _TMA_DT_UINT32, numpy.dtype(numpy.int32): _TMA_DT_INT32, numpy.dtype(numpy.uint64): _TMA_DT_UINT64, numpy.dtype(numpy.int64): _TMA_DT_INT64, numpy.dtype(numpy.float16): _TMA_DT_FLOAT16, numpy.dtype(numpy.float32): _TMA_DT_FLOAT32, numpy.dtype(numpy.float64): _TMA_DT_FLOAT64}
-_TMA_DATA_TYPE_SIZE = {_TMA_DT_UINT8: 1, _TMA_DT_UINT16: 2, _TMA_DT_UINT32: 4, _TMA_DT_INT32: 4, _TMA_DT_UINT64: 8, _TMA_DT_INT64: 8, _TMA_DT_FLOAT16: 2, _TMA_DT_FLOAT32: 4, _TMA_DT_FLOAT64: 8, _TMA_DT_BFLOAT16: 2, _TMA_DT_FLOAT32_FTZ: 4, _TMA_DT_TFLOAT32: 4, _TMA_DT_TFLOAT32_FTZ: 4}
-
-def _normalize_tensor_map_data_type(data_type):
-    ...
-
-def _normalize_tensor_map_sequence(name, values):
-    ...
-
-def _require_tensor_map_enum(name, value, enum_type):
-    ...
-
-def _coerce_tensor_map_descriptor_options(box_dim, options, *, element_strides, data_type, interleave, swizzle, l2_promotion, oob_fill):
-    ...
-
+def _normalize_tensor_map_data_type(data_type): ...
+def _normalize_tensor_map_sequence(name, values): ...
+def _require_tensor_map_enum(name, value, enum_type): ...
+def _coerce_tensor_map_descriptor_options(box_dim, options, *, element_strides, data_type, interleave, swizzle, l2_promotion, oob_fill): ...
 def _resolve_data_type(view, data_type):
     """Resolve the TMA data type from an explicit value or the view's dtype."""
-
 def _get_validated_view(tensor):
     """Obtain a device-accessible StridedMemoryView with a 16-byte-aligned pointer."""
-
 def _require_view_device(view, expected_device_id, operation):
     """Ensure device-local tensors match the current CUDA device.
 
@@ -326,12 +304,10 @@ def _require_view_device(view, expected_device_id, operation):
     ``kDLCUDAManaged`` with ``device_id=0`` regardless of the current device,
     so only true ``kDLCUDA`` tensors are rejected by device-id mismatch.
     """
-
 def _compute_byte_strides(shape, strides, elem_size):
     """Compute byte strides from element strides or C-contiguous fallback.
 
     Returns a tuple of byte strides in row-major order.
     """
-
 def _validate_element_strides(element_strides, rank):
-    """Validate or default element_strides to all-ones."""
\ No newline at end of file
+    """Validate or default element_strides to all-ones."""
diff --git a/cuda_core/cuda/core/_tensor_map.pyx b/cuda_core/cuda/core/_tensor_map.pyx
index 46c2fa93152..3b8b54dd8f3 100644
--- a/cuda_core/cuda/core/_tensor_map.pyx
+++ b/cuda_core/cuda/core/_tensor_map.pyx
@@ -132,19 +132,19 @@ ELSE:
         W128 = 1
 
 
-_TMA_DT_UINT8 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT8)
-_TMA_DT_UINT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT16)
-_TMA_DT_UINT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT32)
-_TMA_DT_INT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT32)
-_TMA_DT_UINT64 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT64)
-_TMA_DT_INT64 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT64)
-_TMA_DT_FLOAT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT16)
-_TMA_DT_FLOAT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32)
-_TMA_DT_FLOAT64 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT64)
-_TMA_DT_BFLOAT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_BFLOAT16)
-_TMA_DT_FLOAT32_FTZ = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32_FTZ)
-_TMA_DT_TFLOAT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32)
-_TMA_DT_TFLOAT32_FTZ = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32_FTZ)
+_TMA_DT_UINT8: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT8)
+_TMA_DT_UINT16: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT16)
+_TMA_DT_UINT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT32)
+_TMA_DT_INT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT32)
+_TMA_DT_UINT64: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT64)
+_TMA_DT_INT64: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_INT64)
+_TMA_DT_FLOAT16: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT16)
+_TMA_DT_FLOAT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32)
+_TMA_DT_FLOAT64: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT64)
+_TMA_DT_BFLOAT16: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_BFLOAT16)
+_TMA_DT_FLOAT32_FTZ: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_FLOAT32_FTZ)
+_TMA_DT_TFLOAT32: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32)
+_TMA_DT_TFLOAT32_FTZ: int = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_TFLOAT32_FTZ)
 
 
 def _normalize_tensor_map_data_type(data_type):
diff --git a/cuda_core/cuda/core/_utils/_weak_handles.pyi b/cuda_core/cuda/core/_utils/_weak_handles.pyi
index 5b7913e008a..a795a180504 100644
--- a/cuda_core/cuda/core/_utils/_weak_handles.pyi
+++ b/cuda_core/cuda/core/_utils/_weak_handles.pyi
@@ -1,4 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_utils/_weak_handles.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_utils/_weak_handles.pyx
 
 """Test-only weak handles for resource-handle lifetime checks.
 
@@ -20,9 +20,6 @@ handle field (see ``*.pxd``), assigns to :ctype:`OpaqueHandle`, and extend the
 Python owners via ``make_opaque_py`` are not covered here -- use
 :class:`weakref.ref` on a weak-referenceable owner object in tests instead.
 """
-from __future__ import annotations
-
-
 class WeakHandle:
     """Non-owning weak handle for a resource's shared control block.
 
@@ -30,13 +27,9 @@ class WeakHandle:
     falsy once the last strong reference is released. Obtain instances via
     :func:`weak_handle` rather than constructing directly.
     """
-
-    def __bool__(self):
-        ...
-
+    def __bool__(self): ...
     def expired(self):
         """Return ``True`` once every strong owner of the handle is gone."""
-
     def use_count(self):
         """Number of strong owners currently sharing the handle."""
 
@@ -53,4 +46,4 @@ def weak_handle(obj):
         If ``obj`` is a :class:`~cuda.core.Buffer` with no active allocation.
     TypeError
         If ``obj`` is not a supported type.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/_utils/_wsl_locale.pyi b/cuda_core/cuda/core/_utils/_wsl_locale.pyi
index 267bdf244f0..790ae71b99d 100644
--- a/cuda_core/cuda/core/_utils/_wsl_locale.pyi
+++ b/cuda_core/cuda/core/_utils/_wsl_locale.pyi
@@ -1,7 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_utils/_wsl_locale.pyx
-
-from __future__ import annotations
-
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_utils/_wsl_locale.pyx
 
 class c_locale_guard:
     """Context manager that pins the calling thread to the "C" locale.
@@ -9,12 +6,6 @@ class c_locale_guard:
     Uses POSIX newlocale/uselocale/freelocale so other threads' view of the
     locale is unaffected. Restores the previous thread locale on exit.
     """
-
-    def __cinit__(self) -> None:
-        ...
-
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc_val, exc_tb):
-        ...
\ No newline at end of file
+    def __init__(self) -> None: ...
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc_val, exc_tb): ...
diff --git a/cuda_core/cuda/core/_utils/cuda_utils.pyi b/cuda_core/cuda/core/_utils/cuda_utils.pyi
index 87067927724..545b9a9073c 100644
--- a/cuda_core/cuda/core/_utils/cuda_utils.pyi
+++ b/cuda_core/cuda/core/_utils/cuda_utils.pyi
@@ -1,21 +1,25 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_utils/cuda_utils.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_utils/cuda_utils.pyx
 
-from __future__ import annotations
-
-from collections import namedtuple
-from typing import Any, Callable
+from typing import Any, Callable, NamedTuple
 
 from cuda.bindings import cydriver
 from cuda.bindings import driver as driver
 from cuda.bindings import nvrtc as nvrtc
 from cuda.bindings import runtime as runtime
 
+_keep_driver_in_stub: driver.CUresult
+_keep_nvrtc_in_stub: nvrtc.nvrtcResult
+_keep_runtime_in_stub: runtime.cudaError_t
+_fork_warning_checked = False
+
+class CUDAError(Exception): ...
 
-class CUDAError(Exception):
-    ...
+class NVRTCError(CUDAError): ...
 
-class NVRTCError(CUDAError):
-    ...
+class ComputeCapability(NamedTuple):
+    """A named tuple of (major, minor) CUDA compute capability version numbers."""
+    major: int
+    minor: int
 
 class Transaction:
     """
@@ -35,81 +39,32 @@ class Transaction:
         append(fn, *args, **kwargs): Register an undo action to be called on rollback.
         commit(): Disarm all undo actions; nothing will be rolled back on exit.
     """
-
-    def __init__(self) -> None:
-        ...
-
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc, tb):
-        ...
-
+    def __init__(self) -> None: ...
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc, tb): ...
     def append(self, fn: Callable[..., Any], /, *args: Any, **kwargs) -> None:
         """
         Register an undo action (runs if the with-block exits without commit()).
         Values are bound now via partial so late mutations don't bite you.
         """
-
     def commit(self) -> None:
         """
         Disarm all undo actions. After this, exiting the with-block does nothing.
         """
-_keep_driver_in_stub: 'driver.CUresult'
-_keep_nvrtc_in_stub: 'nvrtc.nvrtcResult'
-_keep_runtime_in_stub: 'runtime.cudaError_t'
-ComputeCapability = namedtuple('ComputeCapability', ('major', 'minor'))
-_fork_warning_checked = False
-
-def _check_driver_error(error: cydriver.CUresult) -> int:
-    ...
-
-def _check_runtime_error(error) -> int:
-    ...
-
-def _check_nvrtc_error(error, handle=None) -> int:
-    ...
 
+def cast_to_3_tuple(label: str, cfg: int | tuple[int, ...]) -> tuple[int, int, int]: ...
+def _check_driver_error(error: cydriver.CUresult) -> int: ...
+def _check_runtime_error(error) -> int: ...
+def _check_nvrtc_error(error, handle=None) -> int: ...
+def handle_return(result: tuple[Any, ...], handle: object | None=None) -> Any: ...
 def check_or_create_options(cls: type, options: object, options_description: str='', keep_none: bool=False) -> object:
     """
     Create the specified options dataclass from a dictionary of options or None.
     """
-
-def _parse_fill_value(value) -> tuple:
-    """Parse a fill/memset value into (raw_value, element_size).
-
-    Parameters
-    ----------
-    value : int or buffer-protocol object
-        - int: Must be in range [0, 256). Treated as 1-byte fill.
-        - bytes or buffer-protocol: Must be 1, 2, or 4 bytes.
-
-    Returns
-    -------
-    tuple of (int, int)
-        (raw_value, element_size) where element_size is 1, 2, or 4.
-
-    Raises
-    ------
-    OverflowError
-        If int value is outside [0, 256).
-    TypeError
-        If value is not an int and does not support the buffer protocol.
-    ValueError
-        If value byte length is not 1, 2, or 4.
-    """
-
-def cast_to_3_tuple(label: str, cfg: int | tuple[int, ...]) -> tuple[int, int, int]:
-    ...
-
-def handle_return(result: tuple[Any, ...], handle: object=None) -> Any:
-    ...
-
 def _handle_boolean_option(option: bool) -> str:
     """
     Convert a boolean option to a string representation.
     """
-
 def precondition(checker: Callable[..., None], what: str='') -> Callable[..., Any]:
     """
     A decorator that adds checks to ensure any preconditions are met.
@@ -122,23 +77,42 @@ def precondition(checker: Callable[..., None], what: str='') -> Callable[..., An
     Returns:
         Callable: A decorator that creates the wrapping.
     """
-
 def is_sequence(obj: object) -> bool:
     """
     Check if the given object is a sequence (list or tuple).
     """
-
 def is_nested_sequence(obj: object) -> bool:
     """
     Check if the given object is a nested sequence (list or tuple with atleast one list or tuple element).
     """
-
 def reset_fork_warning() -> None:
     """Reset the fork warning check flag for testing purposes.
 
     This function is intended for use in tests to allow multiple test runs
     to check the warning behavior.
     """
+def _parse_fill_value(value) -> tuple[Any, ...]:
+    """Parse a fill/memset value into (raw_value, element_size).
 
+    Parameters
+    ----------
+    value : int or buffer-protocol object
+        - int: Must be in range [0, 256). Treated as 1-byte fill.
+        - bytes or buffer-protocol: Must be 1, 2, or 4 bytes.
+
+    Returns
+    -------
+    tuple of (int, int)
+        (raw_value, element_size) where element_size is 1, 2, or 4.
+
+    Raises
+    ------
+    OverflowError
+        If int value is outside [0, 256).
+    TypeError
+        If value is not an int and does not support the buffer protocol.
+    ValueError
+        If value byte length is not 1, 2, or 4.
+    """
 def check_multiprocessing_start_method() -> None:
-    """Check if multiprocessing start method is 'fork' and warn if so."""
\ No newline at end of file
+    """Check if multiprocessing start method is 'fork' and warn if so."""
diff --git a/cuda_core/cuda/core/_utils/cuda_utils.pyx b/cuda_core/cuda/core/_utils/cuda_utils.pyx
index cf3415b3458..ce75746de56 100644
--- a/cuda_core/cuda/core/_utils/cuda_utils.pyx
+++ b/cuda_core/cuda/core/_utils/cuda_utils.pyx
@@ -7,10 +7,9 @@ from functools import partial
 import multiprocessing
 import platform
 import warnings
-from collections import namedtuple
 from collections.abc import Sequence
 from contextlib import ExitStack
-from typing import Any, Callable
+from typing import Any, Callable, NamedTuple
 
 from cuda.bindings import driver as driver, nvrtc as nvrtc, runtime as runtime
 
@@ -42,7 +41,10 @@ class NVRTCError(CUDAError):
 
 
 
-ComputeCapability = namedtuple("ComputeCapability", ("major", "minor"))
+class ComputeCapability(NamedTuple):
+    """A named tuple of (major, minor) CUDA compute capability version numbers."""
+    major: int
+    minor: int
 
 
 def cast_to_3_tuple(label: str, cfg: int | tuple[int, ...]) -> tuple[int, int, int]:
diff --git a/cuda_core/cuda/core/_utils/version.pyi b/cuda_core/cuda/core/_utils/version.pyi
index a577e037bf7..021d86f1aec 100644
--- a/cuda_core/cuda/core/_utils/version.pyi
+++ b/cuda_core/cuda/core/_utils/version.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/_utils/version.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/_utils/version.pyx
 
 import functools
 
@@ -12,11 +10,9 @@ def _parse_version_triple(version_str: str) -> tuple[int, int, int]:
     ``0b1`` or ``0rc1`` by extracting only the leading integer from each
     release segment.
     """
-
 @functools.cache
 def binding_version() -> tuple[int, int, int]:
     """Return the cuda-bindings version as a (major, minor, patch) triple."""
-
 @functools.cache
 def driver_version() -> tuple[int, int, int]:
-    """Return the CUDA driver version as a (major, minor, patch) triple."""
\ No newline at end of file
+    """Return the CUDA driver version as a (major, minor, patch) triple."""
diff --git a/cuda_core/cuda/core/graph/_adjacency_set_proxy.pyi b/cuda_core/cuda/core/graph/_adjacency_set_proxy.pyi
index 742b3777b04..a434c6d8108 100644
--- a/cuda_core/cuda/core/graph/_adjacency_set_proxy.pyi
+++ b/cuda_core/cuda/core/graph/_adjacency_set_proxy.pyi
@@ -1,8 +1,6 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/graph/_adjacency_set_proxy.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/graph/_adjacency_set_proxy.pyx
 
 """Mutable-set proxy for graph node predecessors and successors."""
-from __future__ import annotations
-
 from collections.abc import Iterator, Set
 from typing import Any
 
@@ -12,47 +10,26 @@ from cuda.core.graph._graph_node import GraphNode
 class AdjacencySetProxy:
     """Mutable set proxy for a node's predecessors or successors.  Mutations
     write through to the underlying CUDA graph."""
-    __slots__ = ('_core',)
-
-    def __init__(self, node: GraphNode, is_fwd: bool) -> None:
-        ...
+    __slots__ = '_core'
 
+    def __init__(self, node: GraphNode, is_fwd: bool) -> None: ...
     @classmethod
-    def _from_iterable(cls, it) -> set[GraphNode]:
-        ...
-
-    def __contains__(self, x: object) -> bool:
-        ...
-
-    def __iter__(self) -> Iterator[GraphNode]:
-        ...
-
-    def __len__(self) -> int:
-        ...
-
-    def add(self, value: GraphNode) -> None:
-        ...
-
-    def discard(self, value: GraphNode) -> None:
-        ...
-
+    def _from_iterable(cls, it) -> set[GraphNode]: ...
+    def __contains__(self, x: object) -> bool: ...
+    def __iter__(self) -> Iterator[GraphNode]: ...
+    def __len__(self) -> int: ...
+    def add(self, value: GraphNode) -> None: ...
+    def discard(self, value: GraphNode) -> None: ...
     def clear(self) -> None:
         """Remove all edges in a single driver call."""
-
-    def __isub__(self, it: Set[Any]) -> 'AdjacencySetProxy':
+    def __isub__(self, it: Set[Any]) -> AdjacencySetProxy:
         """Remove edges to all nodes in *it* in a single driver call."""
-
     def update(self, *others) -> None:
         """Add edges to multiple nodes at once."""
-
-    def __ior__(self, it: Set[Any]) -> 'AdjacencySetProxy':
+    def __ior__(self, it: Set[Any]) -> AdjacencySetProxy:
         """Add edges to all nodes in *it* in a single driver call."""
-
-    def __repr__(self) -> str:
-        ...
+    def __repr__(self) -> str: ...
 
 class _AdjacencySetCore:
     """Cythonized core implementing AdjacencySetProxy"""
-
-    def __init__(self, node: GraphNode, is_fwd: bool):
-        ...
\ No newline at end of file
+    def __init__(self, node: GraphNode, is_fwd: bool): ...
diff --git a/cuda_core/cuda/core/graph/_graph_builder.pyi b/cuda_core/cuda/core/graph/_graph_builder.pyi
index d238b419be1..1e286d3277b 100644
--- a/cuda_core/cuda/core/graph/_graph_builder.pyi
+++ b/cuda_core/cuda/core/graph/_graph_builder.pyi
@@ -1,8 +1,7 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/graph/_graph_builder.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/graph/_graph_builder.pyx
 
 from dataclasses import dataclass
+from typing import TypeAlias
 
 from cuda.core._stream import Stream
 from cuda.core._utils.cuda_utils import driver
@@ -10,8 +9,9 @@ from cuda.core.graph._graph_definition import GraphCondition, GraphDefinition
 from cuda.core.graph._graph_node import GraphNode
 from cuda.core.graph._subclasses import ExecutableGraphNode
 
-_BuilderKind = int
-_CaptureState = int
+_BuilderKind: TypeAlias = int
+_CaptureState: TypeAlias = int
+__all__ = ['Graph', 'GraphBuilder', 'GraphCompleteOptions', 'GraphDebugPrintOptions']
 
 @dataclass
 class GraphDebugPrintOptions:
@@ -121,28 +121,18 @@ class GraphBuilder:
         retains the operands it is given.
 
     """
-
-    def __init__(self):
-        ...
-
-    def __dealloc__(self):
-        ...
-
+    def __init__(self): ...
+    def __dealloc__(self): ...
     @staticmethod
-    def _init(stream: Stream):
-        ...
-
+    def _init(stream: Stream): ...
     def close(self):
         """Destroy the graph builder."""
-
     @property
     def stream(self) -> Stream:
         """Returns the stream associated with the graph builder."""
-
     @property
     def is_join_required(self) -> bool:
         """Returns True if this graph builder must be joined before building is ended."""
-
     @property
     def graph_definition(self) -> GraphDefinition:
         """The captured graph as an explicit :class:`~graph.GraphDefinition`.
@@ -189,7 +179,6 @@ class GraphBuilder:
             keeps working; only fresh access through this property is
             rejected once the builder is closed.
         """
-
     def begin_building(self, mode: str | None='relaxed') -> GraphBuilder:
         """Begins the building process.
 
@@ -206,14 +195,11 @@ class GraphBuilder:
             Default set to use relaxed.
 
         """
-
     @property
     def is_building(self) -> bool:
         """Returns True if the graph builder is currently building."""
-
     def end_building(self) -> GraphBuilder:
         """Ends the building process."""
-
     def complete(self, options: GraphCompleteOptions | None=None) -> Graph:
         """Completes the graph builder and returns the built :obj:`~graph.Graph` object.
 
@@ -228,7 +214,6 @@ class GraphBuilder:
             The newly built graph.
 
         """
-
     def debug_dot_print(self, path: str, options: GraphDebugPrintOptions | None=None) -> None:
         """Generates a DOT debug file for the graph builder.
 
@@ -240,7 +225,6 @@ class GraphBuilder:
             Customizable dataclass for the debug print options.
 
         """
-
     def split(self, count: int) -> tuple[GraphBuilder, ...]:
         """Splits the original graph builder into multiple graph builders.
 
@@ -259,7 +243,6 @@ class GraphBuilder:
             is always the original graph builder.
 
         """
-
     @staticmethod
     def join(*graph_builders: GraphBuilder) -> GraphBuilder:
         """Joins multiple graph builders into a single graph builder.
@@ -277,13 +260,9 @@ class GraphBuilder:
             The newly joined graph builder.
 
         """
-
     def __cuda_stream__(self) -> tuple[int, int]:
         """Return an instance of a __cuda_stream__ protocol."""
-
-    def _get_conditional_context(self) -> driver.CUcontext:
-        ...
-
+    def _get_conditional_context(self) -> driver.CUcontext: ...
     def create_condition(self, default_value: int | None=None) -> GraphCondition:
         """Create a condition variable for use with conditional nodes.
 
@@ -303,7 +282,6 @@ class GraphBuilder:
         GraphCondition
             A condition variable for controlling conditional execution.
         """
-
     def if_then(self, condition: GraphCondition) -> GraphBuilder:
         """Adds an if condition branch and returns a new graph builder for it.
 
@@ -324,7 +302,6 @@ class GraphBuilder:
             The newly created conditional graph builder.
 
         """
-
     def if_else(self, condition: GraphCondition) -> tuple[GraphBuilder, GraphBuilder]:
         """Adds an if-else condition branch and returns new graph builders for both branches.
 
@@ -345,7 +322,6 @@ class GraphBuilder:
             A tuple of two new graph builders, one for the if branch and one for the else branch.
 
         """
-
     def switch(self, condition: GraphCondition, count: int) -> tuple[GraphBuilder, ...]:
         """Adds a switch condition branch and returns new graph builders for all cases.
 
@@ -369,7 +345,6 @@ class GraphBuilder:
             A tuple of new graph builders, one for each branch.
 
         """
-
     def while_loop(self, condition: GraphCondition) -> GraphBuilder:
         """Adds a while loop and returns a new graph builder for it.
 
@@ -390,7 +365,6 @@ class GraphBuilder:
             The newly created while loop graph builder.
 
         """
-
     def embed(self, child: GraphBuilder):
         """Embed a previously-built :obj:`~graph.GraphBuilder` as a child node.
 
@@ -399,7 +373,6 @@ class GraphBuilder:
         child : :obj:`~graph.GraphBuilder`
             The child graph builder. Must have finished building.
         """
-
     def callback(self, fn, *, user_data=None) -> None:
         """Add a host callback to the graph during stream capture.
 
@@ -454,13 +427,9 @@ class Graph:
     Graphs must be built using a :obj:`~graph.GraphBuilder` object.
 
     """
-
-    def __init__(self):
-        ...
-
+    def __init__(self): ...
     def close(self) -> None:
         """Destroy the graph."""
-
     @property
     def handle(self) -> driver.CUgraphExec:
         """Return the underlying ``CUgraphExec`` object.
@@ -471,7 +440,6 @@ class Graph:
             handle, call ``int()`` on the returned object.
 
         """
-
     def __getitem__(self, node: GraphNode) -> ExecutableGraphNode:
         """Return a view for updating *node* in this executable graph.
 
@@ -480,8 +448,7 @@ class Graph:
         node's parameters for future launches. Kernel, memcpy, and memset
         views also support enabling and disabling the node.
         """
-
-    def update(self, source: 'GraphBuilder | GraphDefinition') -> None:
+    def update(self, source: GraphBuilder | GraphDefinition) -> None:
         """Update the graph using a new graph definition.
 
         The topology of the provided source must be identical to this graph.
@@ -493,7 +460,6 @@ class Graph:
             finished building.
 
         """
-
     def upload(self, stream: Stream) -> None:
         """Uploads the graph in a stream.
 
@@ -503,7 +469,6 @@ class Graph:
             The stream in which to upload the graph
 
         """
-
     def launch(self, stream: Stream) -> None:
         """Launches the graph in a stream.
 
@@ -513,10 +478,7 @@ class Graph:
             The stream in which to launch the graph.
 
         """
-__all__ = ['Graph', 'GraphBuilder', 'GraphCompleteOptions', 'GraphDebugPrintOptions']
-
-def _instantiate_graph(source, options: GraphCompleteOptions | None=None) -> Graph:
-    ...
 
+def _instantiate_graph(source, options: GraphCompleteOptions | None=None) -> Graph: ...
 def _capture_callback_with_tail_failure_for_testing(gb: GraphBuilder, fn, *, user_data=None):
-    """Exercise anonymous attachment retention after node discovery fails."""
\ No newline at end of file
+    """Exercise anonymous attachment retention after node discovery fails."""
diff --git a/cuda_core/cuda/core/graph/_graph_definition.pyi b/cuda_core/cuda/core/graph/_graph_definition.pyi
index 9780b53b586..dd915e555a4 100644
--- a/cuda_core/cuda/core/graph/_graph_definition.pyi
+++ b/cuda_core/cuda/core/graph/_graph_definition.pyi
@@ -1,8 +1,6 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/graph/_graph_definition.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/graph/_graph_definition.pyx
 
 """GraphDefinition: explicit CUDA graph definition."""
-from __future__ import annotations
-
 from cuda.core._device import Device
 from cuda.core._event import Event
 from cuda.core._launch_config import LaunchConfig
@@ -20,6 +18,7 @@ from cuda.core.graph._subclasses import (AllocNode, ChildGraphNode, EmptyNode,
                                          WhileNode)
 from cuda.core.typing import GraphMemoryType
 
+__all__ = ['GraphCondition', 'GraphDefinition']
 
 class GraphCondition:
     """A condition variable for conditional graph nodes.
@@ -36,16 +35,9 @@ class GraphCondition:
     ``CUgraphConditionalHandle`` value so device code can update the
     condition.
     """
-
-    def __repr__(self) -> str:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
+    def __repr__(self) -> str: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
     @property
     def handle(self) -> driver.CUgraphConditionalHandle:
         """The raw CUgraphConditionalHandle as an int."""
@@ -63,47 +55,34 @@ class GraphDefinition:
     share underlying graph state. Mutations anywhere in that hierarchy must be
     externally synchronized.
     """
-
     def __init__(self):
         """Create a new empty graph definition."""
-
-    def __repr__(self) -> str:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
+    def __repr__(self) -> str: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
     @property
     def _entry(self) -> GraphNode:
         """Return the internal entry-point GraphNode (no dependencies)."""
-
-    def allocate(self, size: int, *, device: Device | int | None=None, memory_type: GraphMemoryType=..., peer_access: list[Device | int] | None=None) -> AllocNode:
+    def allocate(self, size: int, *, device: Device | int | None=None, memory_type: GraphMemoryType=GraphMemoryType.DEVICE, peer_access: list[Device | int] | None=None) -> AllocNode:
         """Add an entry-point memory allocation node (no dependencies).
 
         See :meth:`GraphNode.allocate` for full documentation.
         """
-
     def deallocate(self, dptr: int) -> FreeNode:
         """Add an entry-point memory free node (no dependencies).
 
         See :meth:`GraphNode.deallocate` for full documentation.
         """
-
     def memset(self, dst: Buffer | int, value, width: int, height: int=1, pitch: int=0, *, dst_owner=None) -> MemsetNode:
         """Add an entry-point memset node (no dependencies).
 
         See :meth:`GraphNode.memset` for full documentation.
         """
-
     def launch(self, config: LaunchConfig, kernel: Kernel, *args) -> KernelNode:
         """Add an entry-point kernel launch node (no dependencies).
 
         See :meth:`GraphNode.launch` for full documentation.
         """
-
     def empty(self) -> EmptyNode:
         """Add an entry-point empty node (no dependencies).
 
@@ -112,7 +91,6 @@ class GraphDefinition:
         EmptyNode
             A new EmptyNode with no dependencies.
         """
-
     def join(self, *nodes: GraphNode) -> EmptyNode:
         """Create an empty node that depends on all given nodes.
 
@@ -126,37 +104,31 @@ class GraphDefinition:
         EmptyNode
             A new EmptyNode that depends on all input nodes.
         """
-
     def memcpy(self, dst: Buffer | int, src: Buffer | int, size: int, *, dst_owner=None, src_owner=None) -> MemcpyNode:
         """Add an entry-point memcpy node (no dependencies).
 
         See :meth:`GraphNode.memcpy` for full documentation.
         """
-
     def embed(self, child: GraphDefinition) -> ChildGraphNode:
         """Add an entry-point child graph node (no dependencies).
 
         See :meth:`GraphNode.embed` for full documentation.
         """
-
     def record(self, event: Event) -> EventRecordNode:
         """Add an entry-point event record node (no dependencies).
 
         See :meth:`GraphNode.record` for full documentation.
         """
-
     def wait(self, event: Event) -> EventWaitNode:
         """Add an entry-point event wait node (no dependencies).
 
         See :meth:`GraphNode.wait` for full documentation.
         """
-
     def callback(self, fn, *, user_data=None) -> HostCallbackNode:
         """Add an entry-point host callback node (no dependencies).
 
         See :meth:`GraphNode.callback` for full documentation.
         """
-
     def create_condition(self, default_value: int | None=None) -> GraphCondition:
         """Create a condition variable for use with conditional nodes.
 
@@ -175,31 +147,26 @@ class GraphDefinition:
         GraphCondition
             A condition variable for controlling conditional execution.
         """
-
     def if_then(self, condition: GraphCondition) -> IfNode:
         """Add an entry-point if-conditional node (no dependencies).
 
         See :meth:`GraphNode.if_then` for full documentation.
         """
-
     def if_else(self, condition: GraphCondition) -> IfElseNode:
         """Add an entry-point if-else conditional node (no dependencies).
 
         See :meth:`GraphNode.if_else` for full documentation.
         """
-
     def while_loop(self, condition: GraphCondition) -> WhileNode:
         """Add an entry-point while-loop conditional node (no dependencies).
 
         See :meth:`GraphNode.while_loop` for full documentation.
         """
-
     def switch(self, condition: GraphCondition, count: int) -> SwitchNode:
         """Add an entry-point switch conditional node (no dependencies).
 
         See :meth:`GraphNode.switch` for full documentation.
         """
-
     def instantiate(self, options: GraphCompleteOptions | None=None) -> Graph:
         """Instantiate the graph definition into an executable Graph.
 
@@ -213,7 +180,6 @@ class GraphDefinition:
         Graph
             An executable graph that can be launched on a stream.
         """
-
     def debug_dot_print(self, path: str, options: GraphDebugPrintOptions | None=None) -> None:
         """Write a GraphViz DOT representation of the graph to a file.
 
@@ -224,7 +190,6 @@ class GraphDefinition:
         options : GraphDebugPrintOptions, optional
             Customizable options for the debug print.
         """
-
     def nodes(self) -> set[GraphNode]:
         """Return all nodes in the graph.
 
@@ -233,7 +198,6 @@ class GraphDefinition:
         set of GraphNode
             All nodes in the graph.
         """
-
     def edges(self) -> set[tuple[GraphNode, GraphNode]]:
         """Return all edges in the graph as (from_node, to_node) pairs.
 
@@ -243,8 +207,6 @@ class GraphDefinition:
             Each element is a (from_node, to_node) pair representing
             a dependency edge in the graph.
         """
-
     @property
     def handle(self) -> driver.CUgraph:
         """Return the underlying driver CUgraph handle."""
-__all__ = ['GraphCondition', 'GraphDefinition']
\ No newline at end of file
diff --git a/cuda_core/cuda/core/graph/_graph_node.pyi b/cuda_core/cuda/core/graph/_graph_node.pyi
index 0e3cac045d2..9c5d9c3a2e0 100644
--- a/cuda_core/cuda/core/graph/_graph_node.pyi
+++ b/cuda_core/cuda/core/graph/_graph_node.pyi
@@ -1,8 +1,6 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/graph/_graph_node.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/graph/_graph_node.pyx
 
 """GraphNode base class — factory, properties, and builder methods."""
-from __future__ import annotations
-
 import weakref
 from collections.abc import Iterable
 
@@ -21,6 +19,8 @@ from cuda.core.graph._subclasses import (AllocNode, ChildGraphNode, EmptyNode,
                                          SwitchNode, WhileNode)
 from cuda.core.typing import GraphMemoryType
 
+__all__ = ['GraphNode']
+_node_registry: weakref.WeakValueDictionary[int, GraphNode] = weakref.WeakValueDictionary()
 
 class GraphNode:
     """A node in a graph definition.
@@ -29,16 +29,9 @@ class GraphNode:
     entry-point nodes with no dependencies) or on other Nodes (for
     nodes that depend on a predecessor).
     """
-
-    def __repr__(self) -> str:
-        ...
-
-    def __eq__(self, other: object) -> bool:
-        ...
-
-    def __hash__(self) -> int:
-        ...
-
+    def __repr__(self) -> str: ...
+    def __eq__(self, other: object) -> bool: ...
+    def __hash__(self) -> int: ...
     @property
     def type(self) -> driver.CUgraphNodeType | None:
         """Return the CUDA graph node type.
@@ -48,25 +41,21 @@ class GraphNode:
         CUgraphNodeType or None
             The node type enum value, or None for the entry node.
         """
-
     @property
     def graph(self) -> GraphDefinition:
         """Return the GraphDefinition this node belongs to."""
-
     @property
     def handle(self) -> driver.CUgraphNode:
         """Return the underlying driver CUgraphNode handle.
 
         Returns None for the entry node.
         """
-
     @property
     def is_valid(self) -> bool:
         """Whether this node is valid (not destroyed).
 
         Returns ``False`` after :meth:`destroy` has been called.
         """
-
     def destroy(self) -> None:
         """Destroy this node and remove all its edges from the parent graph.
 
@@ -74,23 +63,16 @@ class GraphNode:
         cannot be re-added to any graph.  Safe to call on an
         already-destroyed node (no-op).
         """
-
     @property
     def pred(self) -> AdjacencySetProxy:
         """A mutable set-like view of this node's predecessors."""
-
     @pred.setter
-    def pred(self, value: Iterable[GraphNode]) -> None:
-        ...
-
+    def pred(self, value: Iterable[GraphNode]) -> None: ...
     @property
     def succ(self) -> AdjacencySetProxy:
         """A mutable set-like view of this node's successors."""
-
     @succ.setter
-    def succ(self, value: Iterable[GraphNode]) -> None:
-        ...
-
+    def succ(self, value: Iterable[GraphNode]) -> None: ...
     def launch(self, config: LaunchConfig, kernel: Kernel, *args) -> KernelNode:
         """Add a kernel launch node depending on this node.
 
@@ -118,7 +100,6 @@ class GraphNode:
         KernelNode
             A new KernelNode representing the kernel launch.
         """
-
     def join(self, *nodes: GraphNode) -> EmptyNode:
         """Create an empty node that depends on this node and all given nodes.
 
@@ -134,8 +115,7 @@ class GraphNode:
         EmptyNode
             A new EmptyNode that depends on all input nodes.
         """
-
-    def allocate(self, size: int, *, device: Device | int | None=None, memory_type: GraphMemoryType=..., peer_access: list[Device | int] | None=None) -> AllocNode:
+    def allocate(self, size: int, *, device: Device | int | None=None, memory_type: GraphMemoryType=GraphMemoryType.DEVICE, peer_access: list[Device | int] | None=None) -> AllocNode:
         """Add a memory allocation node depending on this node.
 
         Parameters
@@ -174,7 +154,6 @@ class GraphNode:
         IPC (inter-process communication) is not supported for graph
         memory allocation nodes per CUDA documentation.
         """
-
     def deallocate(self, dptr: int) -> FreeNode:
         """Add a memory free node depending on this node.
 
@@ -188,7 +167,6 @@ class GraphNode:
         FreeNode
             A new FreeNode representing the free operation.
         """
-
     def memset(self, dst: Buffer | int, value, width: int, height: int=1, pitch: int=0, *, dst_owner=None) -> MemsetNode:
         """Add a memset node depending on this node.
 
@@ -230,7 +208,6 @@ class GraphNode:
         ValueError
             If ``dst_owner`` is given together with a :class:`Buffer` ``dst``.
         """
-
     def memcpy(self, dst: Buffer | int, src: Buffer | int, size: int, *, dst_owner=None, src_owner=None) -> MemcpyNode:
         """Add a memcpy node depending on this node.
 
@@ -277,7 +254,6 @@ class GraphNode:
             If ``dst_owner`` or ``src_owner`` is given together with a
             :class:`Buffer` ``dst`` or ``src`` respectively.
         """
-
     def embed(self, child: GraphDefinition) -> ChildGraphNode:
         """Add a child graph node depending on this node.
 
@@ -295,7 +271,6 @@ class GraphNode:
         ChildGraphNode
             A new ChildGraphNode representing the embedded sub-graph.
         """
-
     def record(self, event: Event) -> EventRecordNode:
         """Add an event record node depending on this node.
 
@@ -309,7 +284,6 @@ class GraphNode:
         EventRecordNode
             A new EventRecordNode representing the event record operation.
         """
-
     def wait(self, event: Event) -> EventWaitNode:
         """Add an event wait node depending on this node.
 
@@ -323,7 +297,6 @@ class GraphNode:
         EventWaitNode
             A new EventWaitNode representing the event wait operation.
         """
-
     def callback(self, fn, *, user_data=None) -> object:
         """Add a host callback node depending on this node.
 
@@ -372,7 +345,6 @@ class GraphNode:
         ValueError
             If ``user_data`` is given for a Python callable.
         """
-
     def if_then(self, condition: GraphCondition) -> IfNode:
         """Add an if-conditional node depending on this node.
 
@@ -389,7 +361,6 @@ class GraphNode:
         IfNode
             A new IfNode with one branch accessible via ``.then``.
         """
-
     def if_else(self, condition: GraphCondition) -> IfElseNode:
         """Add an if-else conditional node depending on this node.
 
@@ -407,7 +378,6 @@ class GraphNode:
             A new IfElseNode with branches accessible via
             ``.then`` and ``.else_``.
         """
-
     def while_loop(self, condition: GraphCondition) -> WhileNode:
         """Add a while-loop conditional node depending on this node.
 
@@ -424,7 +394,6 @@ class GraphNode:
         WhileNode
             A new WhileNode with body accessible via ``.body``.
         """
-
     def switch(self, condition: GraphCondition, count: int) -> SwitchNode:
         """Add a switch conditional node depending on this node.
 
@@ -443,5 +412,3 @@ class GraphNode:
         SwitchNode
             A new SwitchNode with branches accessible via ``.branches``.
         """
-__all__ = ['GraphNode']
-_node_registry: weakref.WeakValueDictionary[int, GraphNode] = weakref.WeakValueDictionary()
\ No newline at end of file
diff --git a/cuda_core/cuda/core/graph/_host_callback.pyi b/cuda_core/cuda/core/graph/_host_callback.pyi
index 1c642abf501..60674fcedc2 100644
--- a/cuda_core/cuda/core/graph/_host_callback.pyi
+++ b/cuda_core/cuda/core/graph/_host_callback.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/graph/_host_callback.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/graph/_host_callback.pyx
 
 import sys
 
@@ -8,7 +6,6 @@ _CUHOSTFN_HINT = 'ctypes.CFUNCTYPE(None, ctypes.c_void_p)' if sys.platform != 'w
 
 def _cuhostfn_type_error(detail):
     """Build the rejection message for a non-conforming ctypes callback."""
-
 def _validate_ctypes_host_callback(fn):
     """Reject ctypes callbacks whose declared prototype is not CUhostFn.
 
@@ -16,4 +13,4 @@ def _validate_ctypes_host_callback(fn):
     what CUDA calls through. A function pointer taken from a shared library
     keeps ctypes' defaults -- a ``c_int`` result and unspecified arguments --
     until the caller declares otherwise, so it must be declared to be accepted.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/graph/_subclasses.pyi b/cuda_core/cuda/core/graph/_subclasses.pyi
index a68e500f7f2..7a46b2969b8 100644
--- a/cuda_core/cuda/core/graph/_subclasses.pyi
+++ b/cuda_core/cuda/core/graph/_subclasses.pyi
@@ -1,8 +1,6 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/graph/_subclasses.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/graph/_subclasses.pyx
 
 """GraphNode subclasses — EmptyNode through SwitchNode."""
-from __future__ import annotations
-
 from cuda.core._event import Event
 from cuda.core._launch_config import LaunchConfig
 from cuda.core._memory._buffer import Buffer
@@ -11,12 +9,11 @@ from cuda.core.graph._graph_definition import GraphCondition, GraphDefinition
 from cuda.core.graph._graph_node import GraphNode
 from cuda.core.typing import GraphConditionalType
 
+__all__ = ['AllocNode', 'ChildGraphNode', 'ConditionalNode', 'EmptyNode', 'EventRecordNode', 'EventWaitNode', 'ExecutableChildGraphNode', 'ExecutableEventRecordNode', 'ExecutableEventWaitNode', 'ExecutableGraphNode', 'ExecutableHostCallbackNode', 'ExecutableKernelNode', 'ExecutableMemcpyNode', 'ExecutableMemsetNode', 'FreeNode', 'HostCallbackNode', 'IfElseNode', 'IfNode', 'KernelNode', 'MemcpyNode', 'MemsetNode', 'SwitchNode', 'WhileNode']
 
 class EmptyNode(GraphNode):
     """An empty (synchronization) node."""
-
-    def __repr__(self) -> str:
-        ...
+    def __repr__(self) -> str: ...
 
 class KernelNode(GraphNode):
     """A kernel launch node.
@@ -34,10 +31,7 @@ class KernelNode(GraphNode):
     config : LaunchConfig
         A LaunchConfig reconstructed from this node's parameters.
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, *, config: LaunchConfig | None=None, kernel: Kernel | None=None, args=None) -> None:
         """Replace selected kernel launch parameters.
 
@@ -52,23 +46,18 @@ class KernelNode(GraphNode):
             graph that retains it cannot be broken by Python's cyclic garbage
             collector. Use a weak reference to break such cycles.
         """
-
     @property
     def grid(self) -> tuple[int, int, int]:
         """Grid dimensions as a 3-tuple (gridDimX, gridDimY, gridDimZ)."""
-
     @property
     def block(self) -> tuple[int, int, int]:
         """Block dimensions as a 3-tuple (blockDimX, blockDimY, blockDimZ)."""
-
     @property
     def shmem_size(self) -> int:
         """Dynamic shared memory size in bytes."""
-
     @property
     def kernel(self) -> Kernel:
         """The Kernel object for this launch node."""
-
     @property
     def config(self) -> LaunchConfig:
         """A LaunchConfig reconstructed from this node's grid, block, and shmem_size.
@@ -93,26 +82,19 @@ class AllocNode(GraphNode):
     peer_access : tuple of int
         Device IDs that have read-write access to this allocation.
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     @property
     def dptr(self) -> int:
         """The device pointer for the allocation."""
-
     @property
     def bytesize(self) -> int:
         """The number of bytes allocated."""
-
     @property
     def device_id(self) -> int:
         """The device on which the allocation was made."""
-
     @property
     def memory_type(self) -> str:
         """The type of memory: ``"device"``, ``"host"``, or ``"managed"``."""
-
     @property
     def peer_access(self) -> tuple[int, ...]:
         """Device IDs with read-write access to this allocation."""
@@ -125,10 +107,7 @@ class FreeNode(GraphNode):
     dptr : int
         The device pointer being freed.
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     @property
     def dptr(self) -> int:
         """The device pointer being freed."""
@@ -151,10 +130,7 @@ class MemsetNode(GraphNode):
     pitch : int
         Pitch in bytes (unused if height is 1).
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, *, dst: Buffer | int | None=None, value=None, width: int | None=None, height: int | None=None, pitch: int | None=None, dst_owner=None) -> None:
         """Replace selected memset parameters.
 
@@ -172,27 +148,21 @@ class MemsetNode(GraphNode):
             that retains it cannot be broken by Python's cyclic garbage
             collector. Use a weak reference to break such cycles.
         """
-
     @property
     def dptr(self) -> int:
         """The destination device pointer."""
-
     @property
     def value(self) -> int:
         """The fill value."""
-
     @property
     def element_size(self) -> int:
         """Element size in bytes (1, 2, or 4)."""
-
     @property
     def width(self) -> int:
         """Width of the row in elements."""
-
     @property
     def height(self) -> int:
         """Number of rows."""
-
     @property
     def pitch(self) -> int:
         """Pitch in bytes (unused if height is 1)."""
@@ -209,10 +179,7 @@ class MemcpyNode(GraphNode):
     size : int
         The number of bytes copied.
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, *, dst: Buffer | int | None=None, src: Buffer | int | None=None, size: int | None=None, dst_owner=None, src_owner=None) -> None:
         """Replace selected memcpy parameters.
 
@@ -232,15 +199,12 @@ class MemcpyNode(GraphNode):
             that retains it cannot be broken by Python's cyclic garbage
             collector. Use a weak reference to break such cycles.
         """
-
     @property
     def dst(self) -> int:
         """The destination pointer."""
-
     @property
     def src(self) -> int:
         """The source pointer."""
-
     @property
     def size(self) -> int:
         """The number of bytes copied."""
@@ -253,16 +217,12 @@ class ChildGraphNode(GraphNode):
     child_graph : GraphDefinition
         The embedded graph definition (non-owning wrapper).
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, child: GraphDefinition) -> None:
         """Replace the embedded graph with a clone of ``child``.
 
         ``child`` must belong to an independent graph hierarchy.
         """
-
     @property
     def child_graph(self) -> GraphDefinition:
         """The embedded graph definition (non-owning wrapper)."""
@@ -275,13 +235,9 @@ class EventRecordNode(GraphNode):
     event : Event
         The event being recorded.
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, event: Event) -> None:
         """Replace the event recorded by this node."""
-
     @property
     def event(self) -> Event:
         """The event being recorded."""
@@ -294,13 +250,9 @@ class EventWaitNode(GraphNode):
     event : Event
         The event being waited on.
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, event: Event) -> None:
         """Replace the event waited on by this node."""
-
     @property
     def event(self) -> Event:
         """The event being waited on."""
@@ -313,10 +265,7 @@ class HostCallbackNode(GraphNode):
     callback : callable or None
         The Python callable (None for ctypes function pointer callbacks).
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     def update(self, fn, *, user_data=None) -> None:
         """Replace the callback and user-data binding for this node.
 
@@ -335,7 +284,6 @@ class HostCallbackNode(GraphNode):
             retains it cannot be broken by Python's cyclic garbage collector.
             Use a weak reference to break such cycles.
         """
-
     @property
     def callback(self):
         """The Python callable, or None for ctypes function pointer callbacks."""
@@ -358,14 +306,10 @@ class ConditionalNode(GraphNode):
     branches : tuple of GraphDefinition
         The body graphs for each branch (empty pre-13.2).
     """
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     @property
     def condition(self) -> GraphCondition | None:
         """The condition variable controlling execution."""
-
     @property
     def cond_type(self) -> GraphConditionalType | None:
         """The conditional type: GraphConditionalType.IF, .WHILE, or .SWITCH
@@ -373,7 +317,6 @@ class ConditionalNode(GraphNode):
         Returns None when reconstructed from the driver pre-CUDA 13.2,
         as the conditional type cannot be determined.
         """
-
     @property
     def branches(self) -> tuple[GraphDefinition, ...]:
         """The body graphs for each branch as a tuple of GraphDefinition.
@@ -384,43 +327,31 @@ class ConditionalNode(GraphNode):
 
 class IfNode(ConditionalNode):
     """An if-conditional node."""
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     @property
     def then(self) -> GraphDefinition:
         """The 'then' branch graph."""
 
 class IfElseNode(ConditionalNode):
     """An if-else conditional node."""
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     @property
     def then(self) -> GraphDefinition:
         """The ``then`` branch graph (executed when condition is non-zero)."""
-
     @property
     def else_(self) -> GraphDefinition:
         """The ``else`` branch graph (executed when condition is zero)."""
 
 class WhileNode(ConditionalNode):
     """A while-loop conditional node."""
-
-    def __repr__(self) -> str:
-        ...
-
+    def __repr__(self) -> str: ...
     @property
     def body(self) -> GraphDefinition:
         """The loop body graph."""
 
 class SwitchNode(ConditionalNode):
     """A switch conditional node."""
-
-    def __repr__(self) -> str:
-        ...
+    def __repr__(self) -> str: ...
 
 class ExecutableGraphNode:
     """A lightweight view pairing an executable graph with a source node.
@@ -429,16 +360,11 @@ class ExecutableGraphNode:
     node identifies a node in the executable graph when an operation is
     performed.
     """
-
-    def __init__(self):
-        ...
-
-    def __repr__(self) -> str:
-        ...
+    def __init__(self): ...
+    def __repr__(self) -> str: ...
 
 class ExecutableKernelNode(ExecutableGraphNode):
     """An executable kernel-node view."""
-
     def update(self, *, config: LaunchConfig, kernel: Kernel, args) -> None:
         """Replace all kernel launch parameters for future launches.
 
@@ -446,70 +372,55 @@ class ExecutableKernelNode(ExecutableGraphNode):
         for a no-argument kernel. Clustered and cooperative launch
         configurations are not supported.
         """
-
     @property
     def is_enabled(self) -> bool:
         """Whether this node is enabled in the executable graph."""
-
     def enable(self) -> None:
         """Enable this node in the executable graph."""
-
     def disable(self) -> None:
         """Disable this node in the executable graph."""
 
 class ExecutableMemsetNode(ExecutableGraphNode):
     """An executable memset-node view."""
-
     def update(self, *, dst: Buffer | int, value, width: int, height: int=1, pitch: int=0) -> None:
         """Replace all memset parameters for future launches."""
-
     @property
     def is_enabled(self) -> bool:
         """Whether this node is enabled in the executable graph."""
-
     def enable(self) -> None:
         """Enable this node in the executable graph."""
-
     def disable(self) -> None:
         """Disable this node in the executable graph."""
 
 class ExecutableMemcpyNode(ExecutableGraphNode):
     """An executable memcpy-node view."""
-
     def update(self, *, dst: Buffer | int, src: Buffer | int, size: int) -> None:
         """Replace all one-dimensional memcpy parameters for future launches."""
-
     @property
     def is_enabled(self) -> bool:
         """Whether this node is enabled in the executable graph."""
-
     def enable(self) -> None:
         """Enable this node in the executable graph."""
-
     def disable(self) -> None:
         """Disable this node in the executable graph."""
 
 class ExecutableChildGraphNode(ExecutableGraphNode):
     """An executable child-graph-node view."""
-
     def update(self, child: GraphDefinition) -> None:
         """Replace the embedded graph parameters for future launches."""
 
 class ExecutableEventRecordNode(ExecutableGraphNode):
     """An executable event-record-node view."""
-
     def update(self, event: Event) -> None:
         """Replace the event recorded by future launches."""
 
 class ExecutableEventWaitNode(ExecutableGraphNode):
     """An executable event-wait-node view."""
-
     def update(self, event: Event) -> None:
         """Replace the event waited on by future launches."""
 
 class ExecutableHostCallbackNode(ExecutableGraphNode):
     """An executable host-callback-node view."""
-
     def update(self, fn, *, user_data=None) -> None:
         """Replace the callback and user-data binding for future launches.
 
@@ -522,4 +433,3 @@ class ExecutableHostCallbackNode(ExecutableGraphNode):
             Callbacks must not call CUDA API functions. Doing so may deadlock
             or corrupt driver state.
         """
-__all__ = ['AllocNode', 'ChildGraphNode', 'ConditionalNode', 'EmptyNode', 'EventRecordNode', 'EventWaitNode', 'ExecutableChildGraphNode', 'ExecutableEventRecordNode', 'ExecutableEventWaitNode', 'ExecutableGraphNode', 'ExecutableHostCallbackNode', 'ExecutableKernelNode', 'ExecutableMemcpyNode', 'ExecutableMemsetNode', 'FreeNode', 'HostCallbackNode', 'IfElseNode', 'IfNode', 'KernelNode', 'MemcpyNode', 'MemsetNode', 'SwitchNode', 'WhileNode']
\ No newline at end of file
diff --git a/cuda_core/cuda/core/system/_device.pyi b/cuda_core/cuda/core/system/_device.pyi
index c758576f0ac..b2faecc5bb1 100644
--- a/cuda_core/cuda/core/system/_device.pyi
+++ b/cuda_core/cuda/core/system/_device.pyi
@@ -1,8 +1,6 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/system/_device.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/system/_device.pyx
 
-from __future__ import annotations
-
-from typing import Iterable
+from typing import Iterable, TypedDict
 
 import cuda.core
 from cuda.bindings import nvml
@@ -15,27 +13,57 @@ from cuda.core.system.typing import (AddressingMode, AffinityScope, ClockId,
                                      TemperatureThresholds, ThermalController,
                                      ThermalTarget)
 
+_CLOCK_ID_MAPPING = {ClockId.CURRENT: nvml.ClockId.CURRENT, ClockId.CUSTOMER_BOOST_MAX: nvml.ClockId.CUSTOMER_BOOST_MAX}
+_CLOCKS_EVENT_REASONS_MAPPING = {nvml.ClocksEventReasons.EVENT_REASON_NONE: ClocksEventReasons.NONE, nvml.ClocksEventReasons.EVENT_REASON_GPU_IDLE: ClocksEventReasons.GPU_IDLE, nvml.ClocksEventReasons.EVENT_REASON_APPLICATIONS_CLOCKS_SETTING: ClocksEventReasons.APPLICATIONS_CLOCKS_SETTING, nvml.ClocksEventReasons.EVENT_REASON_SW_POWER_CAP: ClocksEventReasons.SW_POWER_CAP, nvml.ClocksEventReasons.THROTTLE_REASON_HW_SLOWDOWN: ClocksEventReasons.HW_SLOWDOWN, nvml.ClocksEventReasons.EVENT_REASON_SYNC_BOOST: ClocksEventReasons.SYNC_BOOST, nvml.ClocksEventReasons.EVENT_REASON_SW_THERMAL_SLOWDOWN: ClocksEventReasons.SW_THERMAL_SLOWDOWN, nvml.ClocksEventReasons.THROTTLE_REASON_HW_THERMAL_SLOWDOWN: ClocksEventReasons.HW_THERMAL_SLOWDOWN, nvml.ClocksEventReasons.THROTTLE_REASON_HW_POWER_BRAKE_SLOWDOWN: ClocksEventReasons.HW_POWER_BRAKE_SLOWDOWN, nvml.ClocksEventReasons.EVENT_REASON_DISPLAY_CLOCK_SETTING: ClocksEventReasons.DISPLAY_CLOCK_SETTING}
+_CLOCK_TYPE_MAPPING = {ClockType.GRAPHICS: nvml.ClockType.CLOCK_GRAPHICS, ClockType.SM: nvml.ClockType.CLOCK_SM, ClockType.MEMORY: nvml.ClockType.CLOCK_MEM, ClockType.VIDEO: nvml.ClockType.CLOCK_VIDEO}
+_COOLER_CONTROL_MAPPING = {nvml.CoolerControl.THERMAL_COOLER_SIGNAL_TOGGLE: CoolerControl.TOGGLE, nvml.CoolerControl.THERMAL_COOLER_SIGNAL_VARIABLE: CoolerControl.VARIABLE}
+_COOLER_TARGET_MAPPING = {nvml.CoolerTarget.THERMAL_NONE: CoolerTarget.NONE, nvml.CoolerTarget.THERMAL_GPU: CoolerTarget.GPU, nvml.CoolerTarget.THERMAL_MEMORY: CoolerTarget.MEMORY, nvml.CoolerTarget.THERMAL_POWER_SUPPLY: CoolerTarget.POWER_SUPPLY}
+_EVENT_TYPE_MAPPING = {nvml.EventType.NONE: EventType.NONE, nvml.EventType.SINGLE_BIT_ECC_ERROR: EventType.SINGLE_BIT_ECC_ERROR, nvml.EventType.DOUBLE_BIT_ECC_ERROR: EventType.DOUBLE_BIT_ECC_ERROR, nvml.EventType.PSTATE: EventType.PSTATE, nvml.EventType.XID_CRITICAL_ERROR: EventType.XID_CRITICAL_ERROR, nvml.EventType.CLOCK: EventType.CLOCK, nvml.EventType.POWER_SOURCE_CHANGE: EventType.POWER_SOURCE_CHANGE, nvml.EventType.MIG_CONFIG_CHANGE: EventType.MIG_CONFIG_CHANGE, nvml.EventType.SINGLE_BIT_ECC_ERROR_STORM: EventType.SINGLE_BIT_ECC_ERROR_STORM, nvml.EventType.DRAM_RETIREMENT_EVENT: EventType.DRAM_RETIREMENT_EVENT, nvml.EventType.DRAM_RETIREMENT_FAILURE: EventType.DRAM_RETIREMENT_FAILURE, nvml.EventType.NON_FATAL_POISON_ERROR: EventType.NON_FATAL_POISON_ERROR, nvml.EventType.FATAL_POISON_ERROR: EventType.FATAL_POISON_ERROR, nvml.EventType.GPU_UNAVAILABLE_ERROR: EventType.GPU_UNAVAILABLE_ERROR, nvml.EventType.GPU_RECOVERY_ACTION: EventType.GPU_RECOVERY_ACTION}
+_EVENT_TYPE_INV_MAPPING = ...
+_FAN_CONTROL_POLICY_MAPPING = {nvml.FanControlPolicy.TEMPERATURE_CONTINUOUS_SW: FanControlPolicy.TEMPERATURE_CONTROLLED, nvml.FanControlPolicy.MANUAL: FanControlPolicy.MANUAL}
+_INFOROM_OBJECT_MAPPING = {InforomObject.OEM: nvml.InforomObject.INFOROM_OEM, InforomObject.ECC: nvml.InforomObject.INFOROM_ECC, InforomObject.POWER: nvml.InforomObject.INFOROM_POWER, InforomObject.DEN: nvml.InforomObject.INFOROM_DEN}
+_NVLINK_VERSION_MAPPING = {nvml.NvlinkVersion.VERSION_1_0: (1, 0), nvml.NvlinkVersion.VERSION_2_0: (2, 0), nvml.NvlinkVersion.VERSION_2_2: (2, 2), nvml.NvlinkVersion.VERSION_3_0: (3, 0), nvml.NvlinkVersion.VERSION_3_1: (3, 1), nvml.NvlinkVersion.VERSION_4_0: (4, 0), nvml.NvlinkVersion.VERSION_5_0: (5, 0)}
+_TEMPERATURE_THRESHOLD_MAPPING = {TemperatureThresholds.SHUTDOWN: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_SHUTDOWN, TemperatureThresholds.SLOWDOWN: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_SLOWDOWN, TemperatureThresholds.MEM_MAX: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_MEM_MAX, TemperatureThresholds.GPU_MAX: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_GPU_MAX, TemperatureThresholds.ACOUSTIC_MIN: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_ACOUSTIC_MIN, TemperatureThresholds.ACOUSTIC_CURR: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_ACOUSTIC_CURR, TemperatureThresholds.ACOUSTIC_MAX: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_ACOUSTIC_MAX, TemperatureThresholds.GPS_CURR: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_GPS_CURR}
+_THERMAL_CONTROLLER_MAPPING = {nvml.ThermalController.GPU_INTERNAL: ThermalController.GPU_INTERNAL, nvml.ThermalController.ADM1032: ThermalController.ADM1032, nvml.ThermalController.ADT7461: ThermalController.ADT7461, nvml.ThermalController.MAX6649: ThermalController.MAX6649, nvml.ThermalController.MAX1617: ThermalController.MAX1617, nvml.ThermalController.LM99: ThermalController.LM99, nvml.ThermalController.LM89: ThermalController.LM89, nvml.ThermalController.LM64: ThermalController.LM64, nvml.ThermalController.G781: ThermalController.G781, nvml.ThermalController.ADT7473: ThermalController.ADT7473, nvml.ThermalController.SBMAX6649: ThermalController.SBMAX6649, nvml.ThermalController.VBIOSEVT: ThermalController.VBIOSEVT, nvml.ThermalController.OS: ThermalController.OS, nvml.ThermalController.NVSYSCON_CANOAS: ThermalController.NVSYSCON_CANOAS, nvml.ThermalController.NVSYSCON_E551: ThermalController.NVSYSCON_E551, nvml.ThermalController.MAX6649R: ThermalController.MAX6649R, nvml.ThermalController.ADT7473S: ThermalController.ADT7473S, nvml.ThermalController.UNKNOWN: ThermalController.UNKNOWN}
+_THERMAL_TARGET_MAPPING = {nvml.ThermalTarget.NONE: ThermalTarget.NONE, nvml.ThermalTarget.GPU: ThermalTarget.GPU, nvml.ThermalTarget.MEMORY: ThermalTarget.MEMORY, nvml.ThermalTarget.POWER_SUPPLY: ThermalTarget.POWER_SUPPLY, nvml.ThermalTarget.BOARD: ThermalTarget.BOARD, nvml.ThermalTarget.VCD_BOARD: ThermalTarget.VCD_BOARD, nvml.ThermalTarget.VCD_INLET: ThermalTarget.VCD_INLET, nvml.ThermalTarget.VCD_OUTLET: ThermalTarget.VCD_OUTLET, nvml.ThermalTarget.ALL: ThermalTarget.ALL}
+_THERMAL_TARGET_INV_MAPPING = ...
+_ADDRESSING_MODE_MAPPING = {nvml.DeviceAddressingModeType.DEVICE_ADDRESSING_MODE_HMM: AddressingMode.HMM, nvml.DeviceAddressingModeType.DEVICE_ADDRESSING_MODE_ATS: AddressingMode.ATS}
+_AFFINITY_SCOPE_MAPPING = {AffinityScope.NODE: nvml.AffinityScope.NODE, AffinityScope.SOCKET: nvml.AffinityScope.SOCKET}
+_BRAND_TYPE_MAPPING = {nvml.BrandType.BRAND_UNKNOWN: 'Unknown', nvml.BrandType.BRAND_QUADRO: 'Quadro', nvml.BrandType.BRAND_TESLA: 'Tesla', nvml.BrandType.BRAND_NVS: 'NVS', nvml.BrandType.BRAND_GRID: 'GRID', nvml.BrandType.BRAND_GEFORCE: 'GeForce', nvml.BrandType.BRAND_TITAN: 'Titan', nvml.BrandType.BRAND_NVIDIA_VAPPS: 'NVIDIA vApps', nvml.BrandType.BRAND_NVIDIA_VPC: 'NVIDIA VPC', nvml.BrandType.BRAND_NVIDIA_VCS: 'NVIDIA VCS', nvml.BrandType.BRAND_NVIDIA_VWS: 'NVIDIA VWS', nvml.BrandType.BRAND_NVIDIA_CLOUD_GAMING: 'NVIDIA Cloud Gaming', nvml.BrandType.BRAND_NVIDIA_VGAMING: 'NVIDIA vGaming', nvml.BrandType.BRAND_QUADRO_RTX: 'Quadro RTX', nvml.BrandType.BRAND_NVIDIA_RTX: 'NVIDIA RTX', nvml.BrandType.BRAND_NVIDIA: 'NVIDIA', nvml.BrandType.BRAND_GEFORCE_RTX: 'GeForce RTX', nvml.BrandType.BRAND_TITAN_RTX: 'Titan RTX'}
+_GPU_P2P_CAPS_INDEX_MAPPING = {GpuP2PCapsIndex.READ: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_READ, GpuP2PCapsIndex.WRITE: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_WRITE, GpuP2PCapsIndex.NVLINK: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_NVLINK, GpuP2PCapsIndex.ATOMICS: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_ATOMICS, GpuP2PCapsIndex.PCI: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_PCI, GpuP2PCapsIndex.PROP: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_PROP, GpuP2PCapsIndex.UNKNOWN: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_UNKNOWN}
+_GPU_P2P_STATUS_MAPPING = {nvml.GpuP2PStatus.P2P_STATUS_OK: GpuP2PStatus.OK, nvml.GpuP2PStatus.P2P_STATUS_CHIPSET_NOT_SUPPORTED: GpuP2PStatus.CHIPSET_NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_GPU_NOT_SUPPORTED: GpuP2PStatus.GPU_NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_IOH_TOPOLOGY_NOT_SUPPORTED: GpuP2PStatus.IOH_TOPOLOGY_NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_DISABLED_BY_REGKEY: GpuP2PStatus.DISABLED_BY_REGKEY, nvml.GpuP2PStatus.P2P_STATUS_NOT_SUPPORTED: GpuP2PStatus.NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_UNKNOWN: GpuP2PStatus.UNKNOWN}
+_GPU_TOPOLOGY_LEVEL_MAPPING = {GpuTopologyLevel.INTERNAL: nvml.GpuTopologyLevel.TOPOLOGY_INTERNAL, GpuTopologyLevel.SINGLE: nvml.GpuTopologyLevel.TOPOLOGY_SINGLE, GpuTopologyLevel.MULTIPLE: nvml.GpuTopologyLevel.TOPOLOGY_MULTIPLE, GpuTopologyLevel.HOSTBRIDGE: nvml.GpuTopologyLevel.TOPOLOGY_HOSTBRIDGE, GpuTopologyLevel.NODE: nvml.GpuTopologyLevel.TOPOLOGY_NODE, GpuTopologyLevel.SYSTEM: nvml.GpuTopologyLevel.TOPOLOGY_SYSTEM}
+_GPU_TOPOLOGY_LEVEL_INV_MAPPING = ...
+__all__ = ['Device', 'get_p2p_status', 'get_topology_common_ancestor', 'NvlinkInfo']
+
+class _GpuDynamicPstatesUtilization(TypedDict):
+    bIsPresent: int
+    percentage: int
+    incThreshold: int
+    decThreshold: int
+
+class _ThermalSensor(TypedDict):
+    controller: int
+    defaultMinTemp: int
+    defaultMaxTemp: int
+    currentTemp: int
+    target: int
 
 class ClockOffsets:
     """
     Contains clock offset information.
     """
-
-    def __init__(self, clock_offset: nvml.ClockOffset):
-        ...
-
+    def __init__(self, clock_offset: nvml.ClockOffset): ...
     @property
     def clock_offset_mhz(self) -> int:
         """
         The current clock offset in MHz.
         """
-
     @property
     def max_offset_mhz(self) -> int:
         """
         The maximum clock offset in MHz.
         """
-
     @property
     def min_offset_mhz(self) -> int:
         """
@@ -46,11 +74,8 @@ class ClockInfo:
     """
     Accesses various clock information about a device.
     """
-
-    def __init__(self, handle: int, clock_type: ClockType | str):
-        ...
-
-    def get_current_mhz(self, clock_id: ClockId | str=...) -> int:
+    def __init__(self, handle: int, clock_type: ClockType | str): ...
+    def get_current_mhz(self, clock_id: ClockId | str=ClockId.CURRENT) -> int:
         """
         Get the current clock speed of a specific clock domain, in MHz.
 
@@ -66,7 +91,6 @@ class ClockInfo:
         int
             The clock speed in MHz.
         """
-
     def get_max_mhz(self) -> int:
         """
         Get the maximum clock speed of a specific clock domain, in MHz.
@@ -81,7 +105,6 @@ class ClockInfo:
         int
             The maximum clock speed in MHz.
         """
-
     def get_max_customer_boost_mhz(self) -> int:
         """
         Get the maximum customer boost clock speed of a specific clock, in MHz.
@@ -93,7 +116,6 @@ class ClockInfo:
         int
             The maximum customer boost clock speed in MHz.
         """
-
     def get_min_max_clock_of_pstate_mhz(self, pstate: int) -> tuple[int, int]:
         """
         Get the minimum and maximum clock speeds for this clock domain
@@ -111,7 +133,6 @@ class ClockInfo:
         tuple[int, int]
             A tuple containing the minimum and maximum clock speeds in MHz.
         """
-
     def get_offsets(self, pstate: int) -> ClockOffsets:
         """
         Retrieve min, max and current clock offset of some clock domain for a given Pstate.
@@ -132,10 +153,7 @@ class ClockInfo:
         """
 
 class CoolerInfo:
-
-    def __init__(self, cooler_info: nvml.CoolerInfo):
-        ...
-
+    def __init__(self, cooler_info: nvml.CoolerInfo): ...
     @property
     def signal_type(self) -> CoolerControl | None:
         """
@@ -143,7 +161,6 @@ class CoolerInfo:
 
         The possible types are variable and toggle.
         """
-
     @property
     def target(self) -> list[CoolerTarget]:
         """
@@ -157,58 +174,47 @@ class DeviceAttributes:
     """
     Various device attributes.
     """
-
-    def __init__(self, attributes: nvml.DeviceAttributes):
-        ...
-
+    def __init__(self, attributes: nvml.DeviceAttributes): ...
     @property
     def multiprocessor_count(self) -> int:
         """
         The streaming multiprocessor count
         """
-
     @property
     def shared_copy_engine_count(self) -> int:
         """
         The shared copy engine count
         """
-
     @property
     def shared_decoder_count(self) -> int:
         """
         The shared decoder engine count
         """
-
     @property
     def shared_encoder_count(self) -> int:
         """
         The shared encoder engine count
         """
-
     @property
     def shared_jpeg_count(self) -> int:
         """
         The shared JPEG engine count
         """
-
     @property
     def shared_ofa_count(self) -> int:
         """
         The shared optical flow accelerator (OFA) engine count
         """
-
     @property
     def gpu_instance_slice_count(self) -> int:
         """
         The GPU instance slice count
         """
-
     @property
     def compute_instance_slice_count(self) -> int:
         """
         The compute instance slice count
         """
-
     @property
     def memory_size_mb(self) -> int:
         """
@@ -219,22 +225,17 @@ class EventData:
     """
     Data about a single event.
     """
-
-    def __init__(self, event_data: nvml.EventData):
-        ...
-
+    def __init__(self, event_data: nvml.EventData): ...
     @property
     def device(self) -> Device:
         """
         The device on which the event occurred.
         """
-
     @property
     def event_type(self) -> EventType:
         """
         The type of event that was triggered.
         """
-
     @property
     def event_data(self) -> int:
         """
@@ -243,7 +244,6 @@ class EventData:
 
         Raises :class:`ValueError` for other event types.
         """
-
     @property
     def gpu_instance_id(self) -> int:
         """
@@ -253,7 +253,6 @@ class EventData:
 
         Raises :class:`ValueError` for other event types.
         """
-
     @property
     def compute_instance_id(self) -> int:
         """
@@ -268,13 +267,8 @@ class DeviceEvents:
     """
     Represents a set of events that can be waited on for a specific device.
     """
-
-    def __init__(self, device_handle: int, events: EventType | str | list[EventType | str]):
-        ...
-
-    def __dealloc__(self) -> None:
-        ...
-
+    def __init__(self, device_handle: int, events: EventType | str | list[EventType | str]): ...
+    def __dealloc__(self) -> None: ...
     def wait(self, timeout_ms: int=0) -> EventData:
         """
         Wait for events in the event set.
@@ -322,10 +316,7 @@ class FanInfo:
     """
     Manages information related to a specific fan on a specific device.
     """
-
-    def __init__(self, handle: int, fan: int):
-        ...
-
+    def __init__(self, handle: int, fan: int): ...
     @property
     def speed(self) -> int:
         """
@@ -340,11 +331,8 @@ class FanInfo:
         The fan speed is expressed as a percentage of the product's maximum
         noise tolerance fan speed.  This value may exceed 100% in certain cases.
         """
-
     @speed.setter
-    def speed(self, speed: int) -> None:
-        ...
-
+    def speed(self, speed: int) -> None: ...
     @property
     def speed_rpm(self) -> int:
         """
@@ -359,7 +347,6 @@ class FanInfo:
         physically blocked and unable to spin, the output will not match the
         actual fan speed.
         """
-
     @property
     def target_speed(self) -> int:
         """
@@ -375,7 +362,6 @@ class FanInfo:
         The fan speed is expressed as a percentage of the product's maximum
         noise tolerance fan speed.  This value may exceed 100% in certain cases.
         """
-
     @property
     def min_max_speed(self) -> tuple[int, int]:
         """
@@ -388,7 +374,6 @@ class FanInfo:
         tuple[int, int]
             A tuple of (min_speed, max_speed)
         """
-
     @property
     def control_policy(self) -> FanControlPolicy:
         """
@@ -398,7 +383,6 @@ class FanInfo:
 
         For all CUDA-capable discrete products with fans.
         """
-
     def set_default_speed(self) -> None:
         """
         Set the speed of the fan control policy to default.
@@ -412,29 +396,23 @@ class FieldValue:
 
     Use :meth:`Device.get_field_values` to get multiple field values at once.
     """
-
-    def __init__(self, field_value: nvml.FieldValue):
-        ...
-
+    def __init__(self, field_value: nvml.FieldValue): ...
     @property
     def field_id(self) -> FieldId:
         """
         The field ID.
         """
-
     @property
     def scope_id(self) -> int:
         """
         The scope ID.
         """
-
     @property
     def timestamp(self) -> int:
         """
         The CPU timestamp (in microseconds since 1970) at which the value was
         sampled.
         """
-
     @property
     def latency_usec(self) -> int:
         """
@@ -442,7 +420,6 @@ class FieldValue:
         be averaged across several fields that are serviced by the same driver
         call.
         """
-
     @property
     def value(self) -> int | float:
         """
@@ -458,16 +435,9 @@ class FieldValues:
     """
     Container of multiple field values.
     """
-
-    def __init__(self, field_values: nvml.FieldValue):
-        ...
-
-    def __getitem__(self, idx: int) -> FieldValue:
-        ...
-
-    def __len__(self) -> int:
-        ...
-
+    def __init__(self, field_values: nvml.FieldValue): ...
+    def __getitem__(self, idx: int) -> FieldValue: ...
+    def __len__(self) -> int: ...
     def validate(self) -> None:
         """
         Validate that there are no issues in any of the contained field values.
@@ -479,7 +449,6 @@ class FieldValues:
         :class:`cuda.core.system.NvmlError`
             If any of the contained field values has an associated exception.
         """
-
     def get_all_values(self) -> list[int | float]:
         """
         Get all field values as a list.
@@ -499,10 +468,7 @@ class FieldValues:
         """
 
 class InforomInfo:
-
-    def __init__(self, device: Device):
-        ...
-
+    def __init__(self, device: Device): ...
     def get_version(self, inforom: InforomObject | str) -> str:
         """
         Retrieves the InfoROM version for a given InfoROM object.
@@ -522,7 +488,6 @@ class InforomInfo:
         str
             The InfoROM version.
         """
-
     @property
     def image_version(self) -> str:
         """
@@ -539,7 +504,6 @@ class InforomInfo:
         str
             The InfoROM image version.
         """
-
     @property
     def configuration_checksum(self) -> int:
         """
@@ -557,7 +521,6 @@ class InforomInfo:
         int
             The InfoROM checksum.
         """
-
     def validate(self) -> None:
         """
         Reads the InfoROM from the flash and verifies the checksums.
@@ -569,7 +532,6 @@ class InforomInfo:
         :class:`cuda.core.system.CorruptedInforomError`
             If the device's InfoROM is corrupted.
         """
-
     @property
     def bbx_flush_time(self) -> tuple[int, int]:
         """
@@ -584,7 +546,6 @@ class InforomInfo:
             - timestamp: The start timestamp of the last BBX flush
             - duration_us: The duration (in μs) of the last BBX flush
         """
-
     @property
     def board_part_number(self) -> str:
         """
@@ -595,28 +556,22 @@ class MemoryInfo:
     """
     Memory allocation information for a device.
     """
-
-    def __init__(self, memory_info: nvml.Memory_v2):
-        ...
-
+    def __init__(self, memory_info: nvml.Memory_v2): ...
     @property
     def free(self) -> int:
         """
         Unallocated device memory (in bytes)
         """
-
     @property
     def total(self) -> int:
         """
         Total physical device memory (in bytes)
         """
-
     @property
     def used(self) -> int:
         """
         Allocated device memory (in bytes)
         """
-
     @property
     def reserved(self) -> int:
         """
@@ -627,22 +582,17 @@ class BAR1MemoryInfo(MemoryInfo):
     """
     BAR1 Memory allocation information for a device.
     """
-
-    def __init__(self, memory_info: nvml.BAR1Memory):
-        ...
-
+    def __init__(self, memory_info: nvml.BAR1Memory): ...
     @property
     def free(self) -> int:
         """
         Unallocated BAR1 memory (in bytes)
         """
-
     @property
     def total(self) -> int:
         """
         Total BAR1 memory (in bytes)
         """
-
     @property
     def used(self) -> int:
         """
@@ -650,10 +600,7 @@ class BAR1MemoryInfo(MemoryInfo):
         """
 
 class MigInfo:
-
-    def __init__(self, device: Device):
-        ...
-
+    def __init__(self, device: Device): ...
     @property
     def is_mig_device(self) -> bool:
         """
@@ -666,7 +613,6 @@ class MigInfo:
 
         For Ampere™ or newer fully supported devices.
         """
-
     @property
     def mode(self) -> bool:
         """
@@ -682,7 +628,6 @@ class MigInfo:
         bool
             `True` if current MIG mode is enabled.
         """
-
     @mode.setter
     def mode(self, mode: bool) -> None:
         """
@@ -698,7 +643,6 @@ class MigInfo:
         mode: bool
             `True` to enable MIG mode, `False` to disable MIG mode.
         """
-
     @property
     def pending_mode(self) -> bool:
         """
@@ -716,7 +660,6 @@ class MigInfo:
         bool
             `True` if pending MIG mode is enabled.
         """
-
     @property
     def device_count(self) -> int:
         """
@@ -731,7 +674,6 @@ class MigInfo:
         int
             The number of MIG devices (compute instances) on this GPU.
         """
-
     @property
     def parent(self) -> Device:
         """
@@ -744,7 +686,6 @@ class MigInfo:
         Device
             The parent GPU device for this MIG device.
         """
-
     def get_device_by_index(self, index: int) -> Device:
         """
         Get MIG device for the given index under its parent device.
@@ -768,7 +709,6 @@ class MigInfo:
         Device
             The MIG device corresponding to the given index.
         """
-
     def get_all_devices(self) -> Iterable[Device]:
         """
         Get all MIG devices under its parent device.
@@ -788,7 +728,6 @@ class MigInfo:
         """
 
 class _NvlinkInfoMeta(type):
-
     @property
     def max_links(cls):
         """
@@ -807,10 +746,7 @@ class _NvlinkInfo:
     """
     Nvlink information for a device.
     """
-
-    def __init__(self, device: Device, link: int):
-        ...
-
+    def __init__(self, device: Device, link: int): ...
     @property
     def version(self) -> tuple[int, int]:
         """
@@ -823,7 +759,6 @@ class _NvlinkInfo:
         tuple[int, int]
             The Nvlink version as a tuple of (major, minor).
         """
-
     @property
     def state(self) -> bool:
         """
@@ -839,71 +774,58 @@ class _NvlinkInfo:
             `True` if the Nvlink is active.
         """
 
-class NvlinkInfo(_NvlinkInfo, metaclass=_NvlinkInfoMeta):
-    ...
+class NvlinkInfo(_NvlinkInfo, metaclass=_NvlinkInfoMeta): ...
 
 class PciInfo:
     """
     PCI information about a GPU device.
     """
-
-    def __init__(self, pci_info_ext: nvml.PciInfoExt_v1, handle: int):
-        ...
-
+    def __init__(self, pci_info_ext: nvml.PciInfoExt_v1, handle: int): ...
     @property
     def bus(self) -> int:
         """
         The bus on which the device resides, 0 to 255
         """
-
     @property
     def bus_id(self) -> str:
         """
         The tuple domain:bus:device.function PCI identifier string
         """
-
     @property
     def device(self) -> int:
         """
         The device's id on the bus, 0 to 31
         """
-
     @property
     def domain(self) -> int:
         """
         The PCI domain on which the device's bus resides, 0 to 0xffffffff
         """
-
     @property
     def vendor_id(self) -> int:
         """
         The PCI vendor id of the device
         """
-
     @property
     def device_id(self) -> int:
         """
         The PCI device id of the device
         """
-
     @property
     def subsystem_id(self) -> int:
         """
         The subsystem device ID
         """
-
     @property
     def base_class(self) -> int:
         """
         The 8-bit PCI base class code
         """
-
     @property
     def sub_class(self) -> int:
         """
         The 8-bit PCI sub class code
         """
-
     @property
     def link_generation(self) -> int:
         """
@@ -915,7 +837,6 @@ class PciInfo:
         PCIe bus, the max link generation this function will report is
         generation 1.
         """
-
     @property
     def max_link_generation(self) -> int:
         """
@@ -923,7 +844,6 @@ class PciInfo:
 
         For Fermi™ or newer fully supported devices.
         """
-
     @property
     def max_link_width(self) -> int:
         """
@@ -935,7 +855,6 @@ class PciInfo:
         PCIe system bus this function will report
         a max link width of 8.
         """
-
     @property
     def current_link_generation(self) -> int:
         """
@@ -943,7 +862,6 @@ class PciInfo:
 
         For Fermi™ or newer fully supported devices.
         """
-
     @property
     def current_link_width(self) -> int:
         """
@@ -951,7 +869,6 @@ class PciInfo:
 
         For Fermi™ or newer fully supported devices.
         """
-
     @property
     def rx_throughput(self) -> int:
         """
@@ -965,7 +882,6 @@ class PciInfo:
         This method is not supported in virtual machines running virtual GPU
         (vGPU).
         """
-
     @property
     def tx_throughput(self) -> int:
         """
@@ -979,7 +895,6 @@ class PciInfo:
         This method is not supported in virtual machines running virtual GPU
         (vGPU).
         """
-
     @property
     def replay_counter(self) -> int:
         """
@@ -989,28 +904,22 @@ class PciInfo:
         """
 
 class GpuDynamicPstatesUtilization:
-
-    def __init__(self, ptr: int, owner: object):
-        ...
-
+    def __init__(self, ptr: int, owner: object): ...
     @property
     def is_present(self) -> bool:
         """
         Set if the utilization domain is present on this GPU.
         """
-
     @property
     def percentage(self) -> int:
         """
         Percentage of time where the domain is considered busy in the last 1-second interval.
         """
-
     @property
     def inc_threshold(self) -> int:
         """
         Utilization threshold that can trigger a perf-increasing P-State change when crossed.
         """
-
     @property
     def dec_threshold(self) -> int:
         """
@@ -1021,36 +930,25 @@ class GpuDynamicPstatesInfo:
     """
     Handles performance monitor samples from the device.
     """
-
-    def __init__(self, gpu_dynamic_pstates_info: nvml.GpuDynamicPstatesInfo):
-        ...
-
-    def __len__(self) -> int:
-        ...
-
-    def __getitem__(self, idx: int) -> GpuDynamicPstatesUtilization:
-        ...
+    def __init__(self, gpu_dynamic_pstates_info: nvml.GpuDynamicPstatesInfo): ...
+    def __len__(self) -> int: ...
+    def __getitem__(self, idx: int) -> GpuDynamicPstatesUtilization: ...
 
 class ProcessInfo:
     """
     Information about running compute processes on the GPU.
     """
-
-    def __init__(self, device: Device, process_info: nvml.ProcessInfo):
-        ...
-
+    def __init__(self, device: Device, process_info: nvml.ProcessInfo): ...
     @property
     def pid(self) -> int:
         """
         The PID of the process.
         """
-
     @property
     def used_gpu_memory(self) -> int:
         """
         The amount of GPU memory (in bytes) used by the process.
         """
-
     @property
     def gpu_instance_id(self) -> int:
         """
@@ -1058,7 +956,6 @@ class ProcessInfo:
 
         Only valid for processes running on MIG devices.
         """
-
     @property
     def compute_instance_id(self) -> int:
         """
@@ -1071,16 +968,12 @@ class RepairStatus:
     """
     Repair status for TPC/Channel repair.
     """
-
-    def __init__(self, handle: int):
-        ...
-
+    def __init__(self, handle: int): ...
     @property
     def channel_repair_pending(self) -> bool:
         """
         `True` if a channel repair is pending.
         """
-
     @property
     def tpc_repair_pending(self) -> bool:
         """
@@ -1088,46 +981,25 @@ class RepairStatus:
         """
 
 class ThermalSensor:
-
-    def __init__(self, ptr: int, owner: object):
-        ...
-
+    def __init__(self, ptr: int, owner: object): ...
     @property
-    def controller(self) -> ThermalController:
-        ...
-
+    def controller(self) -> ThermalController: ...
     @property
-    def default_min_temp(self) -> int:
-        ...
-
+    def default_min_temp(self) -> int: ...
     @property
-    def default_max_temp(self) -> int:
-        ...
-
+    def default_max_temp(self) -> int: ...
     @property
-    def current_temp(self) -> int:
-        ...
-
+    def current_temp(self) -> int: ...
     @property
-    def target(self) -> ThermalTarget:
-        ...
+    def target(self) -> ThermalTarget: ...
 
 class ThermalSettings:
-
-    def __init__(self, thermal_settings: nvml.ThermalSettings):
-        ...
-
-    def __len__(self) -> int:
-        ...
-
-    def __getitem__(self, idx: int) -> nvml.ThermalSensor:
-        ...
+    def __init__(self, thermal_settings: nvml.ThermalSettings): ...
+    def __len__(self) -> int: ...
+    def __getitem__(self, idx: int) -> nvml.ThermalSensor: ...
 
 class Temperature:
-
-    def __init__(self, handle: int):
-        ...
-
+    def __init__(self, handle: int): ...
     def get_sensor(self) -> int:
         """
         Get the temperature reading from a specific sensor on the device, in
@@ -1140,7 +1012,6 @@ class Temperature:
         int
             The temperature in degrees Celsius.
         """
-
     def get_threshold(self, threshold_type: TemperatureThresholds | str) -> int:
         """
         Retrieves the temperature threshold for this GPU with the specified
@@ -1162,13 +1033,11 @@ class Temperature:
         use :meth:`get_field_values` with ``NVML_FI_DEV_TEMPERATURE_*`` fields
         to retrieve temperature thresholds on these architectures.
         """
-
     @property
     def margin(self) -> int:
         """
         The thermal margin temperature (distance to nearest slowdown threshold) for the device.
         """
-
     def get_thermal_settings(self, sensor_index: ThermalTarget | str) -> ThermalSettings:
         """
         Used to execute a list of thermal system instructions.
@@ -1190,16 +1059,12 @@ class Utilization:
 
     For devices with compute capability 2.0 or higher.
     """
-
-    def __init__(self, utilization: nvml.Utilization):
-        ...
-
+    def __init__(self, utilization: nvml.Utilization): ...
     @property
     def gpu(self) -> int:
         """
         Percent of time over the past sample period during which one or more kernels was executing on the GPU.
         """
-
     @property
     def memory(self) -> int:
         """
@@ -1240,9 +1105,7 @@ class Device:
     """
     _handle: int
 
-    def __init__(self, *, index: int | None=None, uuid: bytes | str | None=None, pci_bus_id: bytes | str | None=None) -> None:
-        ...
-
+    def __init__(self, *, index: int | None=None, uuid: bytes | str | None=None, pci_bus_id: bytes | str | None=None) -> None: ...
     @property
     def index(self) -> int:
         """
@@ -1260,7 +1123,6 @@ class Device:
         Note: The NVML index may not correlate with other APIs, such as the CUDA
         device index.
         """
-
     @property
     def uuid(self) -> str:
         """
@@ -1272,7 +1134,6 @@ class Device:
         prefix.  If you need a `uuid` without that prefix (for example, to
         interact with CUDA), use the `uuid_without_prefix` property.
         """
-
     @property
     def uuid_without_prefix(self) -> str:
         """
@@ -1284,13 +1145,11 @@ class Device:
         prefix.  This property returns it without the prefix, to match the UUIDs
         used in CUDA.  If you need the prefix, use the `uuid` property.
         """
-
     @property
     def pci_bus_id(self) -> str:
         """
         Retrieves the PCI bus ID of this device.
         """
-
     @property
     def numa_node_id(self) -> int:
         """
@@ -1298,7 +1157,6 @@ class Device:
 
         This only applies to platforms where the GPUs are NUMA nodes.
         """
-
     @property
     def arch(self) -> DeviceArch:
         """
@@ -1308,13 +1166,11 @@ class Device:
         "VOLTA"``, and RTX A6000 will report ``DeviceArchitecture.name ==
         "AMPERE"``.
         """
-
     @property
     def name(self) -> str:
         """
         Name of the device, e.g.: `"Tesla V100-SXM2-32GB"`
         """
-
     @property
     def brand(self) -> str:
         """
@@ -1322,7 +1178,6 @@ class Device:
 
         Returns "Unknown" if the brand is unknown.
         """
-
     @property
     def serial(self) -> str:
         """
@@ -1331,7 +1186,6 @@ class Device:
 
         For all products with an InfoROM.
         """
-
     @property
     def module_id(self) -> int:
         """
@@ -1341,7 +1195,6 @@ class Device:
         on a given baseboard.  For non-baseboard products, this ID would always
         be 0.
         """
-
     @property
     def minor_number(self) -> int:
         """
@@ -1352,13 +1205,11 @@ class Device:
         The minor number is used by the Linux device driver to identify the
         device node in ``/dev/nvidiaX``.
         """
-
     @property
     def is_c2c_enabled(self) -> bool:
         """
         Whether the C2C (Chip-to-Chip) mode is enabled for this device.
         """
-
     @property
     def is_persistence_mode_enabled(self) -> bool:
         """
@@ -1366,11 +1217,8 @@ class Device:
 
         For Linux only.
         """
-
     @is_persistence_mode_enabled.setter
-    def is_persistence_mode_enabled(self, enabled: bool) -> None:
-        ...
-
+    def is_persistence_mode_enabled(self, enabled: bool) -> None: ...
     @property
     def cuda_compute_capability(self) -> tuple[int, int]:
         """
@@ -1378,8 +1226,7 @@ class Device:
 
         Returns a tuple `(major, minor)`.
         """
-
-    def to_cuda_device(self) -> 'cuda.core.Device':
+    def to_cuda_device(self) -> cuda.core.Device:
         """
         Get the corresponding :class:`cuda.core.Device` (which is used for CUDA
         access) for this :class:`cuda.core.system.Device` (which is used for
@@ -1401,7 +1248,6 @@ class Device:
             available CUDA device, since it can not be used directly, even
             though it can be enumerated from NVML.
         """
-
     @classmethod
     def get_device_count(cls) -> int:
         """
@@ -1412,7 +1258,6 @@ class Device:
         int
             The number of available devices.
         """
-
     @classmethod
     def get_all_devices(cls) -> Iterable[Device]:
         """
@@ -1423,13 +1268,11 @@ class Device:
         Iterator over :obj:`~Device`
             An iterator over available devices.
         """
-
     @property
     def addressing_mode(self) -> AddressingMode | None:
         """
         Get the :obj:`~AddressingMode` of the device.
         """
-
     @property
     def mig(self) -> MigInfo:
         """
@@ -1437,7 +1280,6 @@ class Device:
 
         For Ampere™ or newer fully supported devices.
         """
-
     @classmethod
     def get_all_devices_with_cpu_affinity(cls, cpu_index: int) -> Iterable[Device]:
         """
@@ -1455,8 +1297,7 @@ class Device:
         Iterator of :obj:`~Device`
             An iterator over available devices.
         """
-
-    def get_memory_affinity(self, scope: AffinityScope | str=...) -> list[int]:
+    def get_memory_affinity(self, scope: AffinityScope | str=AffinityScope.NODE) -> list[int]:
         """
         Retrieves a list of indices of NUMA nodes or CPU sockets with the ideal
         memory affinity for the device.
@@ -1481,8 +1322,7 @@ class Device:
             A list of indices of NUMA nodes or CPU sockets with the ideal memory
             affinity for the device.
         """
-
-    def get_cpu_affinity(self, scope: AffinityScope | str=...) -> list[int]:
+    def get_cpu_affinity(self, scope: AffinityScope | str=AffinityScope.NODE) -> list[int]:
         """
         Retrieves a list of indices of NUMA nodes or CPU sockets with the ideal
         CPU affinity for the device.
@@ -1507,7 +1347,6 @@ class Device:
             A list of indices of NUMA nodes or CPU sockets with the ideal memory
             affinity for the device.
         """
-
     def set_cpu_affinity(self) -> None:
         """
         Sets the ideal affinity for the calling thread and device.
@@ -1516,7 +1355,6 @@ class Device:
 
         Supported on Linux only.
         """
-
     def clear_cpu_affinity(self) -> None:
         """
         Clear all affinity bindings for the calling thread.
@@ -1525,12 +1363,10 @@ class Device:
 
         Supported on Linux only.
         """
-
     def get_clock(self, clock_type: ClockType | str) -> ClockInfo:
         """
         :obj:`~_device.ClockInfo` object to get information about and manage a specific clock on a device.
         """
-
     @property
     def is_auto_boosted_clocks_enabled(self) -> tuple[bool, bool]:
         """
@@ -1554,7 +1390,6 @@ class Device:
             The default Auto Boosted clocks behavior
 
         """
-
     @property
     def current_clock_event_reasons(self) -> list[ClocksEventReasons]:
         """
@@ -1562,7 +1397,6 @@ class Device:
 
         For all fully supported products.
         """
-
     @property
     def supported_clock_event_reasons(self) -> list[ClocksEventReasons]:
         """
@@ -1573,13 +1407,11 @@ class Device:
 
         This method is not supported in virtual machines running virtual GPU (vGPU).
         """
-
     @property
     def cooler(self) -> CoolerInfo:
         """
         :obj:`~_device.CoolerInfo` object with cooler information for the device.
         """
-
     @property
     def attributes(self) -> DeviceAttributes:
         """
@@ -1588,7 +1420,6 @@ class Device:
         For Ampere™ or newer fully supported devices.  Only available on Linux
         systems.
         """
-
     @property
     def is_display_connected(self) -> bool:
         """
@@ -1597,7 +1428,6 @@ class Device:
         Indicates whether a physical display (e.g. monitor) is currently connected to
         any of the device's connectors.
         """
-
     @property
     def is_display_active(self) -> bool:
         """
@@ -1609,7 +1439,6 @@ class Device:
 
         Display can be active even when no monitor is physically attached.
         """
-
     def register_events(self, events: EventType | str | list[EventType | str]) -> DeviceEvents:
         """
         Starts recording events on this device.
@@ -1650,7 +1479,6 @@ class Device:
         :class:`cuda.core.system.NotSupportedError`
             None of the requested event types are registered.
         """
-
     def get_supported_event_types(self) -> list[EventType]:
         """
         Get the list of event types supported by this device.
@@ -1663,18 +1491,15 @@ class Device:
         list[EventType]
             The list of supported event types.
         """
-
     def get_fan(self, fan: int=0) -> FanInfo:
         """
         :obj:`~_device.FanInfo` object to get information and manage a specific fan on a device.
         """
-
     @property
     def num_fans(self) -> int:
         """
         The number of fans on the device.
         """
-
     def get_field_values(self, field_ids: list[int | tuple[int, int]]) -> FieldValues:
         """
         Get multiple field values from the device.
@@ -1699,7 +1524,6 @@ class Device:
         :obj:`~_device.FieldValues`
             Container of field values corresponding to the requested field IDs.
         """
-
     def clear_field_values(self, field_ids: list[int | tuple[int, int]]) -> None:
         """
         Clear multiple field values from the device.
@@ -1712,7 +1536,6 @@ class Device:
             Each item may be either a single value from the :class:`FieldId`
             enum, or a pair of (:class:`FieldId`, scope ID).
         """
-
     @property
     def inforom(self) -> InforomInfo:
         """
@@ -1720,7 +1543,6 @@ class Device:
 
         For all products with an InfoROM.
         """
-
     @property
     def bar1_memory_info(self) -> BAR1MemoryInfo:
         """
@@ -1730,13 +1552,11 @@ class Device:
         accessed by the CPU or by 3rd party devices (peer-to-peer on the PCIE
         bus).
         """
-
     @property
     def memory_info(self) -> MemoryInfo:
         """
         :obj:`~_device.MemoryInfo` object with memory information.
         """
-
     def get_nvlink(self, link: int) -> NvlinkInfo:
         """
         Get :obj:`~NvlinkInfo` about this device.
@@ -1746,7 +1566,6 @@ class Device:
         .. version-changed:: 1.1.0
             Any link number not supported by this specific device will raise a `ValueError`.
         """
-
     def get_nvlink_count(self) -> int:
         """
         Get the number of NVLink links on this device.
@@ -1755,7 +1574,6 @@ class Device:
 
         .. version-added:: 1.1.0
         """
-
     def get_nvlinks(self) -> Iterable[NvlinkInfo]:
         """
         Get :obj:`~NvlinkInfo` about all NVLink links on this device.
@@ -1764,7 +1582,6 @@ class Device:
 
         .. version-added:: 1.1.0
         """
-
     @property
     def pci_info(self) -> PciInfo:
         """
@@ -1773,7 +1590,6 @@ class Device:
         Non-physical devices, such as MIG devices, may not have PCI attributes.
         In that case, this property will raise a `RuntimeError`.
         """
-
     @property
     def performance_state(self) -> int | None:
         """
@@ -1788,13 +1604,11 @@ class Device:
             where 0 is maximum performance and higher numbers are lower performance.
             Returns `None` if the performance state is unknown.
         """
-
     @property
     def dynamic_pstates_info(self) -> GpuDynamicPstatesInfo:
         """
         :obj:`~_device.GpuDynamicPstatesInfo` object with performance monitor samples from the associated subdevice.
         """
-
     @property
     def supported_pstates(self) -> list[int]:
         """
@@ -1810,7 +1624,6 @@ class Device:
             between 0 and 15, where 0 is maximum performance and higher numbers
             are lower performance.
         """
-
     @property
     def compute_running_processes(self) -> list[ProcessInfo]:
         """
@@ -1832,7 +1645,6 @@ class Device:
         Querying per-instance information using MIG device handles is not
         supported if the device is in vGPU Host virtualization mode.
         """
-
     @property
     def repair_status(self) -> RepairStatus:
         """
@@ -1840,13 +1652,11 @@ class Device:
 
         For Ampere™ or newer fully supported devices.
         """
-
     @property
     def temperature(self) -> Temperature:
         """
         :obj:`~_device.Temperature` object with temperature information for the device.
         """
-
     def get_topology_nearest_gpus(self, level: GpuTopologyLevel | str) -> Iterable[Device]:
         """
         Retrieve the GPUs that are nearest to this device at a specific interconnectivity level.
@@ -1863,7 +1673,6 @@ class Device:
         Iterable of :class:`Device`
             The nearest devices at the given topology level.
         """
-
     @property
     def utilization(self) -> Utilization:
         """
@@ -1884,35 +1693,11 @@ class Device:
         Utilization
             An object containing the current utilization rates for the device.
         """
-_CLOCK_ID_MAPPING = {ClockId.CURRENT: nvml.ClockId.CURRENT, ClockId.CUSTOMER_BOOST_MAX: nvml.ClockId.CUSTOMER_BOOST_MAX}
-_CLOCKS_EVENT_REASONS_MAPPING = {nvml.ClocksEventReasons.EVENT_REASON_NONE: ClocksEventReasons.NONE, nvml.ClocksEventReasons.EVENT_REASON_GPU_IDLE: ClocksEventReasons.GPU_IDLE, nvml.ClocksEventReasons.EVENT_REASON_APPLICATIONS_CLOCKS_SETTING: ClocksEventReasons.APPLICATIONS_CLOCKS_SETTING, nvml.ClocksEventReasons.EVENT_REASON_SW_POWER_CAP: ClocksEventReasons.SW_POWER_CAP, nvml.ClocksEventReasons.THROTTLE_REASON_HW_SLOWDOWN: ClocksEventReasons.HW_SLOWDOWN, nvml.ClocksEventReasons.EVENT_REASON_SYNC_BOOST: ClocksEventReasons.SYNC_BOOST, nvml.ClocksEventReasons.EVENT_REASON_SW_THERMAL_SLOWDOWN: ClocksEventReasons.SW_THERMAL_SLOWDOWN, nvml.ClocksEventReasons.THROTTLE_REASON_HW_THERMAL_SLOWDOWN: ClocksEventReasons.HW_THERMAL_SLOWDOWN, nvml.ClocksEventReasons.THROTTLE_REASON_HW_POWER_BRAKE_SLOWDOWN: ClocksEventReasons.HW_POWER_BRAKE_SLOWDOWN, nvml.ClocksEventReasons.EVENT_REASON_DISPLAY_CLOCK_SETTING: ClocksEventReasons.DISPLAY_CLOCK_SETTING}
-_CLOCK_TYPE_MAPPING = {ClockType.GRAPHICS: nvml.ClockType.CLOCK_GRAPHICS, ClockType.SM: nvml.ClockType.CLOCK_SM, ClockType.MEMORY: nvml.ClockType.CLOCK_MEM, ClockType.VIDEO: nvml.ClockType.CLOCK_VIDEO}
-_COOLER_CONTROL_MAPPING = {nvml.CoolerControl.THERMAL_COOLER_SIGNAL_TOGGLE: CoolerControl.TOGGLE, nvml.CoolerControl.THERMAL_COOLER_SIGNAL_VARIABLE: CoolerControl.VARIABLE}
-_COOLER_TARGET_MAPPING = {nvml.CoolerTarget.THERMAL_NONE: CoolerTarget.NONE, nvml.CoolerTarget.THERMAL_GPU: CoolerTarget.GPU, nvml.CoolerTarget.THERMAL_MEMORY: CoolerTarget.MEMORY, nvml.CoolerTarget.THERMAL_POWER_SUPPLY: CoolerTarget.POWER_SUPPLY}
-_EVENT_TYPE_MAPPING = {nvml.EventType.NONE: EventType.NONE, nvml.EventType.SINGLE_BIT_ECC_ERROR: EventType.SINGLE_BIT_ECC_ERROR, nvml.EventType.DOUBLE_BIT_ECC_ERROR: EventType.DOUBLE_BIT_ECC_ERROR, nvml.EventType.PSTATE: EventType.PSTATE, nvml.EventType.XID_CRITICAL_ERROR: EventType.XID_CRITICAL_ERROR, nvml.EventType.CLOCK: EventType.CLOCK, nvml.EventType.POWER_SOURCE_CHANGE: EventType.POWER_SOURCE_CHANGE, nvml.EventType.MIG_CONFIG_CHANGE: EventType.MIG_CONFIG_CHANGE, nvml.EventType.SINGLE_BIT_ECC_ERROR_STORM: EventType.SINGLE_BIT_ECC_ERROR_STORM, nvml.EventType.DRAM_RETIREMENT_EVENT: EventType.DRAM_RETIREMENT_EVENT, nvml.EventType.DRAM_RETIREMENT_FAILURE: EventType.DRAM_RETIREMENT_FAILURE, nvml.EventType.NON_FATAL_POISON_ERROR: EventType.NON_FATAL_POISON_ERROR, nvml.EventType.FATAL_POISON_ERROR: EventType.FATAL_POISON_ERROR, nvml.EventType.GPU_UNAVAILABLE_ERROR: EventType.GPU_UNAVAILABLE_ERROR, nvml.EventType.GPU_RECOVERY_ACTION: EventType.GPU_RECOVERY_ACTION}
-_EVENT_TYPE_INV_MAPPING = {v: k for k, v in _EVENT_TYPE_MAPPING.items()}
-_FAN_CONTROL_POLICY_MAPPING = {nvml.FanControlPolicy.TEMPERATURE_CONTINUOUS_SW: FanControlPolicy.TEMPERATURE_CONTROLLED, nvml.FanControlPolicy.MANUAL: FanControlPolicy.MANUAL}
-_INFOROM_OBJECT_MAPPING = {InforomObject.OEM: nvml.InforomObject.INFOROM_OEM, InforomObject.ECC: nvml.InforomObject.INFOROM_ECC, InforomObject.POWER: nvml.InforomObject.INFOROM_POWER, InforomObject.DEN: nvml.InforomObject.INFOROM_DEN}
-_NVLINK_VERSION_MAPPING = {nvml.NvlinkVersion.VERSION_1_0: (1, 0), nvml.NvlinkVersion.VERSION_2_0: (2, 0), nvml.NvlinkVersion.VERSION_2_2: (2, 2), nvml.NvlinkVersion.VERSION_3_0: (3, 0), nvml.NvlinkVersion.VERSION_3_1: (3, 1), nvml.NvlinkVersion.VERSION_4_0: (4, 0), nvml.NvlinkVersion.VERSION_5_0: (5, 0)}
-_NVLINK_VERSION_6_0 = getattr(nvml.NvlinkVersion, 'VERSION_6_0', None)
-_TEMPERATURE_THRESHOLD_MAPPING = {TemperatureThresholds.SHUTDOWN: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_SHUTDOWN, TemperatureThresholds.SLOWDOWN: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_SLOWDOWN, TemperatureThresholds.MEM_MAX: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_MEM_MAX, TemperatureThresholds.GPU_MAX: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_GPU_MAX, TemperatureThresholds.ACOUSTIC_MIN: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_ACOUSTIC_MIN, TemperatureThresholds.ACOUSTIC_CURR: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_ACOUSTIC_CURR, TemperatureThresholds.ACOUSTIC_MAX: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_ACOUSTIC_MAX, TemperatureThresholds.GPS_CURR: nvml.TemperatureThresholds.TEMPERATURE_THRESHOLD_GPS_CURR}
-_THERMAL_CONTROLLER_MAPPING = {nvml.ThermalController.GPU_INTERNAL: ThermalController.GPU_INTERNAL, nvml.ThermalController.ADM1032: ThermalController.ADM1032, nvml.ThermalController.ADT7461: ThermalController.ADT7461, nvml.ThermalController.MAX6649: ThermalController.MAX6649, nvml.ThermalController.MAX1617: ThermalController.MAX1617, nvml.ThermalController.LM99: ThermalController.LM99, nvml.ThermalController.LM89: ThermalController.LM89, nvml.ThermalController.LM64: ThermalController.LM64, nvml.ThermalController.G781: ThermalController.G781, nvml.ThermalController.ADT7473: ThermalController.ADT7473, nvml.ThermalController.SBMAX6649: ThermalController.SBMAX6649, nvml.ThermalController.VBIOSEVT: ThermalController.VBIOSEVT, nvml.ThermalController.OS: ThermalController.OS, nvml.ThermalController.NVSYSCON_CANOAS: ThermalController.NVSYSCON_CANOAS, nvml.ThermalController.NVSYSCON_E551: ThermalController.NVSYSCON_E551, nvml.ThermalController.MAX6649R: ThermalController.MAX6649R, nvml.ThermalController.ADT7473S: ThermalController.ADT7473S, nvml.ThermalController.UNKNOWN: ThermalController.UNKNOWN}
-_THERMAL_TARGET_MAPPING = {nvml.ThermalTarget.NONE: ThermalTarget.NONE, nvml.ThermalTarget.GPU: ThermalTarget.GPU, nvml.ThermalTarget.MEMORY: ThermalTarget.MEMORY, nvml.ThermalTarget.POWER_SUPPLY: ThermalTarget.POWER_SUPPLY, nvml.ThermalTarget.BOARD: ThermalTarget.BOARD, nvml.ThermalTarget.VCD_BOARD: ThermalTarget.VCD_BOARD, nvml.ThermalTarget.VCD_INLET: ThermalTarget.VCD_INLET, nvml.ThermalTarget.VCD_OUTLET: ThermalTarget.VCD_OUTLET, nvml.ThermalTarget.ALL: ThermalTarget.ALL}
-_THERMAL_TARGET_INV_MAPPING = {v: k for k, v in _THERMAL_TARGET_MAPPING.items()}
-_ADDRESSING_MODE_MAPPING = {nvml.DeviceAddressingModeType.DEVICE_ADDRESSING_MODE_HMM: AddressingMode.HMM, nvml.DeviceAddressingModeType.DEVICE_ADDRESSING_MODE_ATS: AddressingMode.ATS}
-_AFFINITY_SCOPE_MAPPING = {AffinityScope.NODE: nvml.AffinityScope.NODE, AffinityScope.SOCKET: nvml.AffinityScope.SOCKET}
-_BRAND_TYPE_MAPPING = {nvml.BrandType.BRAND_UNKNOWN: 'Unknown', nvml.BrandType.BRAND_QUADRO: 'Quadro', nvml.BrandType.BRAND_TESLA: 'Tesla', nvml.BrandType.BRAND_NVS: 'NVS', nvml.BrandType.BRAND_GRID: 'GRID', nvml.BrandType.BRAND_GEFORCE: 'GeForce', nvml.BrandType.BRAND_TITAN: 'Titan', nvml.BrandType.BRAND_NVIDIA_VAPPS: 'NVIDIA vApps', nvml.BrandType.BRAND_NVIDIA_VPC: 'NVIDIA VPC', nvml.BrandType.BRAND_NVIDIA_VCS: 'NVIDIA VCS', nvml.BrandType.BRAND_NVIDIA_VWS: 'NVIDIA VWS', nvml.BrandType.BRAND_NVIDIA_CLOUD_GAMING: 'NVIDIA Cloud Gaming', nvml.BrandType.BRAND_NVIDIA_VGAMING: 'NVIDIA vGaming', nvml.BrandType.BRAND_QUADRO_RTX: 'Quadro RTX', nvml.BrandType.BRAND_NVIDIA_RTX: 'NVIDIA RTX', nvml.BrandType.BRAND_NVIDIA: 'NVIDIA', nvml.BrandType.BRAND_GEFORCE_RTX: 'GeForce RTX', nvml.BrandType.BRAND_TITAN_RTX: 'Titan RTX'}
-_GPU_P2P_CAPS_INDEX_MAPPING = {GpuP2PCapsIndex.READ: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_READ, GpuP2PCapsIndex.WRITE: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_WRITE, GpuP2PCapsIndex.NVLINK: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_NVLINK, GpuP2PCapsIndex.ATOMICS: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_ATOMICS, GpuP2PCapsIndex.PCI: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_PCI, GpuP2PCapsIndex.PROP: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_PROP, GpuP2PCapsIndex.UNKNOWN: nvml.GpuP2PCapsIndex.P2P_CAPS_INDEX_UNKNOWN}
-_GPU_P2P_STATUS_MAPPING = {nvml.GpuP2PStatus.P2P_STATUS_OK: GpuP2PStatus.OK, nvml.GpuP2PStatus.P2P_STATUS_CHIPSET_NOT_SUPPORTED: GpuP2PStatus.CHIPSET_NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_GPU_NOT_SUPPORTED: GpuP2PStatus.GPU_NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_IOH_TOPOLOGY_NOT_SUPPORTED: GpuP2PStatus.IOH_TOPOLOGY_NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_DISABLED_BY_REGKEY: GpuP2PStatus.DISABLED_BY_REGKEY, nvml.GpuP2PStatus.P2P_STATUS_NOT_SUPPORTED: GpuP2PStatus.NOT_SUPPORTED, nvml.GpuP2PStatus.P2P_STATUS_UNKNOWN: GpuP2PStatus.UNKNOWN}
-_GPU_TOPOLOGY_LEVEL_MAPPING = {GpuTopologyLevel.INTERNAL: nvml.GpuTopologyLevel.TOPOLOGY_INTERNAL, GpuTopologyLevel.SINGLE: nvml.GpuTopologyLevel.TOPOLOGY_SINGLE, GpuTopologyLevel.MULTIPLE: nvml.GpuTopologyLevel.TOPOLOGY_MULTIPLE, GpuTopologyLevel.HOSTBRIDGE: nvml.GpuTopologyLevel.TOPOLOGY_HOSTBRIDGE, GpuTopologyLevel.NODE: nvml.GpuTopologyLevel.TOPOLOGY_NODE, GpuTopologyLevel.SYSTEM: nvml.GpuTopologyLevel.TOPOLOGY_SYSTEM}
-_GPU_TOPOLOGY_LEVEL_INV_MAPPING = {v: k for k, v in _GPU_TOPOLOGY_LEVEL_MAPPING.items()}
-__all__ = ['Device', 'get_p2p_status', 'get_topology_common_ancestor', 'NvlinkInfo']
 
 def _unpack_bitmask(arr: object) -> list[int]:
     """
     Unpack a list of integers containing bitmasks.
     """
-
 def get_topology_common_ancestor(device1: Device, device2: Device) -> GpuTopologyLevel:
     """
     Retrieve the common ancestor for two devices.
@@ -1931,7 +1716,6 @@ def get_topology_common_ancestor(device1: Device, device2: Device) -> GpuTopolog
     :class:`GpuTopologyLevel`
         The common ancestor level of the two devices.
     """
-
 def get_p2p_status(device1: Device, device2: Device, index: GpuP2PCapsIndex | str) -> GpuP2PStatus:
     """
     Retrieve the P2P status between two devices.
@@ -1949,4 +1733,4 @@ def get_p2p_status(device1: Device, device2: Device, index: GpuP2PCapsIndex | st
     -------
     :class:`GpuP2PStatus`
         The P2P status between the two devices.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/system/_nvml_context.pyi b/cuda_core/cuda/core/system/_nvml_context.pyi
index e52a803b346..d61f31ddf40 100644
--- a/cuda_core/cuda/core/system/_nvml_context.pyi
+++ b/cuda_core/cuda/core/system/_nvml_context.pyi
@@ -1,17 +1,14 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/system/_nvml_context.pyx
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/system/_nvml_context.pyx
 
-from __future__ import annotations
+from typing import TypeAlias
 
-import threading
-
-_NVMLState = int
-_lock = threading.Lock()
+_NVMLState: TypeAlias = int
 
+def _get_nvml_state() -> _NVMLState: ...
 def _initialize() -> None:
     """
     Initializes NVIDIA Management Library (NVML), ensuring it only happens once per process.
     """
-
 def validate() -> None:
     """
     Validate NVML state.
@@ -28,6 +25,3 @@ def validate() -> None:
     nvml.GpuNotFoundError
         If no GPUs are available.
     """
-
-def _get_nvml_state() -> _NVMLState:
-    ...
\ No newline at end of file
diff --git a/cuda_core/cuda/core/system/_system.pyi b/cuda_core/cuda/core/system/_system.pyi
index f25ce35be7f..0584101fa96 100644
--- a/cuda_core/cuda/core/system/_system.pyi
+++ b/cuda_core/cuda/core/system/_system.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/system/_system.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/system/_system.pyx
 
 CUDA_BINDINGS_NVML_IS_COMPATIBLE: bool
 __all__ = ['get_driver_branch', 'get_kernel_mode_driver_version', 'get_user_mode_driver_version', 'get_nvml_version', 'get_num_devices', 'get_process_name', 'CUDA_BINDINGS_NVML_IS_COMPATIBLE']
@@ -17,7 +15,6 @@ def get_user_mode_driver_version() -> tuple[int, ...]:
     version : tuple[int, ...]
         A 2-tuple ``(MAJOR, MINOR)``, e.g. ``(13, 0)`` for CUDA 13.0.
     """
-
 def get_kernel_mode_driver_version() -> tuple[int, ...]:
     """
     Get the kernel-mode (KMD / GPU) driver version, e.g. 580.65.06.
@@ -33,7 +30,6 @@ def get_kernel_mode_driver_version() -> tuple[int, ...]:
     RuntimeError
         If the NVML library is not available.
     """
-
 def get_nvml_version() -> tuple[int, ...]:
     """
     The version of the NVML library.
@@ -43,7 +39,6 @@ def get_nvml_version() -> tuple[int, ...]:
     version: tuple[int, ...]
         Tuple of integers representing the NVML version components.
     """
-
 def get_driver_branch() -> str:
     """
     Retrieves the driver branch of the NVIDIA driver installed on the system.
@@ -53,12 +48,10 @@ def get_driver_branch() -> str:
     branch: str
         The driver branch string (e.g., ``"560"``, ``"open"``, etc.).
     """
-
 def get_num_devices() -> int:
     """
     Return the number of devices in the system.
     """
-
 def get_process_name(pid: int) -> str:
     """
     The name of process with given PID.
@@ -72,4 +65,4 @@ def get_process_name(pid: int) -> str:
     -------
     name: str
         The process name.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/system/_system_events.pyi b/cuda_core/cuda/core/system/_system_events.pyi
index 5ae5b86bc57..ef367602975 100644
--- a/cuda_core/cuda/core/system/_system_events.pyi
+++ b/cuda_core/cuda/core/system/_system_events.pyi
@@ -1,33 +1,29 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/system/_system_events.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/system/_system_events.pyx
 
 from cuda.bindings import nvml
 from cuda.core.system.typing import SystemEventType
 
 from . import _device
 
+_SYSTEM_EVENT_TYPE_MAPPING = {nvml.SystemEventType.GPU_DRIVER_UNBIND: SystemEventType.UNBIND, nvml.SystemEventType.GPU_DRIVER_BIND: SystemEventType.BIND}
+_SYSTEM_EVENT_TYPE_INV_MAPPING = ...
+__all__ = ['register_events']
 
 class SystemEvent:
     """
     Data about a collection of system events.
     """
-
-    def __init__(self, event_data: nvml.SystemEventData_v1):
-        ...
-
+    def __init__(self, event_data: nvml.SystemEventData_v1): ...
     @property
     def event_type(self) -> SystemEventType:
         """
         The :obj:`~SystemEventType` that was triggered.
         """
-
     @property
     def gpu_id(self) -> int:
         """
         The GPU ID in PCI ID format.
         """
-
     @property
     def device(self) -> _device.Device:
         """
@@ -38,13 +34,8 @@ class SystemEvents:
     """
     Data about a collection of system events.
     """
-
-    def __init__(self, event_data: nvml.SystemEventData_v1):
-        ...
-
-    def __len__(self) -> int:
-        ...
-
+    def __init__(self, event_data: nvml.SystemEventData_v1): ...
+    def __len__(self) -> int: ...
     def __getitem__(self, idx: int) -> SystemEvent:
         """
         Get the :obj:`~_system_events.SystemEvent` at the specified index.
@@ -54,13 +45,8 @@ class RegisteredSystemEvents:
     """
     Represents a set of events that can be waited on for a specific device.
     """
-
-    def __init__(self, events: SystemEventType | str | list[SystemEventType | str]):
-        ...
-
-    def __dealloc__(self) -> None:
-        ...
-
+    def __init__(self, events: SystemEventType | str | list[SystemEventType | str]): ...
+    def __dealloc__(self) -> None: ...
     def wait(self, timeout_ms: int=0, buffer_size: int=1) -> SystemEvents:
         """
         Wait for events in the system event set.
@@ -95,9 +81,6 @@ class RegisteredSystemEvents:
         :class:`cuda.core.system.GpuIsLostError`
             If the GPU has fallen off the bus or is otherwise inaccessible.
         """
-_SYSTEM_EVENT_TYPE_MAPPING = {nvml.SystemEventType.GPU_DRIVER_UNBIND: SystemEventType.UNBIND, nvml.SystemEventType.GPU_DRIVER_BIND: SystemEventType.BIND}
-_SYSTEM_EVENT_TYPE_INV_MAPPING = {v: k for k, v in _SYSTEM_EVENT_TYPE_MAPPING.items()}
-__all__ = ['register_events']
 
 def register_events(events: SystemEventType | str | list[SystemEventType | str]) -> RegisteredSystemEvents:
     """
@@ -130,4 +113,4 @@ def register_events(events: SystemEventType | str | list[SystemEventType | str])
     ------
     :class:`cuda.core.system.NotSupportedError`
         None of the requested event types are registered.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/texture/_array.pyi b/cuda_core/cuda/core/texture/_array.pyi
index 380c2fe1c10..87b3530c63a 100644
--- a/cuda_core/cuda/core/texture/_array.pyi
+++ b/cuda_core/cuda/core/texture/_array.pyi
@@ -1,13 +1,14 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/texture/_array.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/texture/_array.pyx
 
 from dataclasses import dataclass
 
-import numpy
 from cuda.bindings import cydriver
 from cuda.core.typing import ArrayFormatType
 
+_ARRAYFORMAT_TO_CU = {ArrayFormatType.UINT8: int(cydriver.CU_AD_FORMAT_UNSIGNED_INT8), ArrayFormatType.UINT16: int(cydriver.CU_AD_FORMAT_UNSIGNED_INT16), ArrayFormatType.UINT32: int(cydriver.CU_AD_FORMAT_UNSIGNED_INT32), ArrayFormatType.INT8: int(cydriver.CU_AD_FORMAT_SIGNED_INT8), ArrayFormatType.INT16: int(cydriver.CU_AD_FORMAT_SIGNED_INT16), ArrayFormatType.INT32: int(cydriver.CU_AD_FORMAT_SIGNED_INT32), ArrayFormatType.FLOAT16: int(cydriver.CU_AD_FORMAT_HALF), ArrayFormatType.FLOAT32: int(cydriver.CU_AD_FORMAT_FLOAT)}
+_CU_TO_ARRAYFORMAT = ...
+_NUMPY_DTYPE_TO_ARRAYFORMAT = ...
+_FORMAT_ELEM_SIZE = {_ARRAYFORMAT_TO_CU[ArrayFormatType.UINT8]: 1, _ARRAYFORMAT_TO_CU[ArrayFormatType.INT8]: 1, _ARRAYFORMAT_TO_CU[ArrayFormatType.UINT16]: 2, _ARRAYFORMAT_TO_CU[ArrayFormatType.INT16]: 2, _ARRAYFORMAT_TO_CU[ArrayFormatType.FLOAT16]: 2, _ARRAYFORMAT_TO_CU[ArrayFormatType.UINT32]: 4, _ARRAYFORMAT_TO_CU[ArrayFormatType.INT32]: 4, _ARRAYFORMAT_TO_CU[ArrayFormatType.FLOAT32]: 4}
 
 @dataclass
 class OpaqueArrayOptions:
@@ -35,8 +36,7 @@ class OpaqueArrayOptions:
     num_channels: int
     is_surface_load_store: bool = False
 
-    def __post_init__(self):
-        ...
+    def __post_init__(self): ...
 
 class OpaqueArray:
     """An opaque, hardware-laid-out GPU allocation for texture/surface access.
@@ -61,19 +61,7 @@ class OpaqueArray:
 
     .. versionadded:: 1.1.0
     """
-
-    def close(self):
-        """Release this object's reference to the underlying ``CUarray``.
-
-        Destruction (``cuArrayDestroy``) happens via the handle's deleter when
-        the last reference is dropped; for a non-owning handle (graphics interop
-        or a mipmap-level view) nothing is destroyed. Idempotent: a second call
-        (or destruction after ``close()``) is a no-op.
-        """
-
-    def __init__(self, *args, **kwargs):
-        ...
-
+    def __init__(self, *args, **kwargs): ...
     @classmethod
     def _from_handle(cls, handle: int, owning: bool, *, device_id=None):
         """Wrap an externally-allocated ``CUarray``.
@@ -83,40 +71,31 @@ class OpaqueArray:
         underlying ``CUarray`` is never destroyed by this object. Shape, format,
         and channel count are queried from the driver.
         """
-
     @property
     def handle(self):
         """The underlying ``CUarray`` as an integer."""
-
     @property
     def shape(self):
         """Allocation shape, in elements."""
-
     @property
     def format(self):
         """The element :class:`~cuda.core.typing.ArrayFormatType`."""
-
     @property
     def num_channels(self):
         """Channels per element (1, 2, or 4)."""
-
     @property
     def element_bytes(self):
         """Bytes per element (format size * channels)."""
-
     @property
     def device(self):
         """The :class:`Device` this array was allocated on."""
-
     @property
     def is_surface_load_store(self):
         """True if this array was created with ``CUDA_ARRAY3D_SURFACE_LDST``
         and can be bound as a :class:`SurfaceObject`."""
-
     def _extent_bytes(self):
         """Return (width_bytes, height, depth) for cuMemcpy3D, with height/depth
         normalized to >=1 for lower-rank arrays."""
-
     def copy_from(self, src, *, stream) -> None:
         """Copy a full-array's worth of data into this array.
 
@@ -129,7 +108,6 @@ class OpaqueArray:
             Stream to issue the copy on. A :class:`~cuda.core.graph.GraphBuilder`
             is accepted so the copy can be captured into a graph.
         """
-
     def copy_to(self, dst, *, stream):
         """Copy a full-array's worth of data out of this array.
 
@@ -146,23 +124,20 @@ class OpaqueArray:
         -------
         The ``dst`` object, for parity with :meth:`Buffer.copy_to`.
         """
-
     @property
     def size_bytes(self):
         """Total bytes of array storage (``prod(shape) * element_bytes``)."""
+    def close(self):
+        """Release this object's reference to the underlying ``CUarray``.
 
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc, tb):
-        ...
-
-    def __repr__(self):
-        ...
-_ARRAYFORMAT_TO_CU = {ArrayFormatType.UINT8: int(cydriver.CU_AD_FORMAT_UNSIGNED_INT8), ArrayFormatType.UINT16: int(cydriver.CU_AD_FORMAT_UNSIGNED_INT16), ArrayFormatType.UINT32: int(cydriver.CU_AD_FORMAT_UNSIGNED_INT32), ArrayFormatType.INT8: int(cydriver.CU_AD_FORMAT_SIGNED_INT8), ArrayFormatType.INT16: int(cydriver.CU_AD_FORMAT_SIGNED_INT16), ArrayFormatType.INT32: int(cydriver.CU_AD_FORMAT_SIGNED_INT32), ArrayFormatType.FLOAT16: int(cydriver.CU_AD_FORMAT_HALF), ArrayFormatType.FLOAT32: int(cydriver.CU_AD_FORMAT_FLOAT)}
-_CU_TO_ARRAYFORMAT = {cu: fmt for fmt, cu in _ARRAYFORMAT_TO_CU.items()}
-_NUMPY_DTYPE_TO_ARRAYFORMAT = {numpy.dtype(fmt.value): fmt for fmt in ArrayFormatType}
-_FORMAT_ELEM_SIZE = {_ARRAYFORMAT_TO_CU[ArrayFormatType.UINT8]: 1, _ARRAYFORMAT_TO_CU[ArrayFormatType.INT8]: 1, _ARRAYFORMAT_TO_CU[ArrayFormatType.UINT16]: 2, _ARRAYFORMAT_TO_CU[ArrayFormatType.INT16]: 2, _ARRAYFORMAT_TO_CU[ArrayFormatType.FLOAT16]: 2, _ARRAYFORMAT_TO_CU[ArrayFormatType.UINT32]: 4, _ARRAYFORMAT_TO_CU[ArrayFormatType.INT32]: 4, _ARRAYFORMAT_TO_CU[ArrayFormatType.FLOAT32]: 4}
+        Destruction (``cuArrayDestroy``) happens via the handle's deleter when
+        the last reference is dropped; for a non-owning handle (graphics interop
+        or a mipmap-level view) nothing is destroyed. Idempotent: a second call
+        (or destruction after ``close()``) is a no-op.
+        """
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc, tb): ...
+    def __repr__(self): ...
 
 def _normalize_array_format(format):
     """Coerce ``format`` to an :class:`ArrayFormatType`.
@@ -176,16 +151,13 @@ def _normalize_array_format(format):
       supported formats.
 
     Raises :class:`ValueError` on anything else."""
-
 def _validate_format_channels(format, num_channels):
     """Validate the ``(format, num_channels)`` pair shared by the array,
     mipmap, and texture factories. Returns the normalized
     :class:`ArrayFormatType`. Raises on an invalid combination."""
-
 def _validate_array_shape(shape):
     """Coerce ``shape`` to a tuple of ints and validate rank (1-3) and that
     every extent is >= 1. Returns the normalized tuple."""
-
 def _create_opaque_array(options):
     """Allocate a new :class:`OpaqueArray` on the current device.
 
@@ -193,4 +165,4 @@ def _create_opaque_array(options):
     :class:`OpaqueArrayOptions` (or a mapping accepted by it); it is validated
     at construction, so ``shape`` is already a normalized tuple and ``format``
     an :class:`~cuda.core.typing.ArrayFormatType`.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/texture/_mipmapped_array.pyi b/cuda_core/cuda/core/texture/_mipmapped_array.pyi
index db4413dbaf4..51835f8b36e 100644
--- a/cuda_core/cuda/core/texture/_mipmapped_array.pyi
+++ b/cuda_core/cuda/core/texture/_mipmapped_array.pyi
@@ -1,6 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/texture/_mipmapped_array.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/texture/_mipmapped_array.pyx
 
 from dataclasses import dataclass
 
@@ -37,8 +35,7 @@ class MipmappedArrayOptions:
     num_levels: int
     is_surface_load_store: bool = False
 
-    def __post_init__(self):
-        ...
+    def __post_init__(self): ...
 
 class MipmappedArray:
     """A mipmapped CUDA array for texture/surface access across levels.
@@ -54,19 +51,7 @@ class MipmappedArray:
 
     .. versionadded:: 1.1.0
     """
-
-    def close(self):
-        """Release this object's reference to the underlying ``CUmipmappedArray``.
-
-        Destruction (``cuMipmappedArrayDestroy``) happens via the handle's
-        deleter when the last reference is dropped. A level :class:`OpaqueArray`
-        from :meth:`get_level` holds its own reference to this mipmap's storage,
-        so it stays valid until both it and this object are released. Idempotent.
-        """
-
-    def __init__(self, *args, **kwargs):
-        ...
-
+    def __init__(self, *args, **kwargs): ...
     def get_level(self, level):
         """Return a non-owning :class:`OpaqueArray` view of the given mip level.
 
@@ -83,44 +68,39 @@ class MipmappedArray:
             returned :class:`OpaqueArray`; the underlying storage is released only
             when this :class:`MipmappedArray` is destroyed.
         """
-
     @property
     def handle(self):
         """The underlying ``CUmipmappedArray`` as an integer."""
-
     @property
     def shape(self):
         """Base-level (level 0) allocation shape, in elements."""
-
     @property
     def format(self):
         """The element :class:`~cuda.core.typing.ArrayFormatType`."""
-
     @property
     def num_channels(self):
         """Channels per element (1, 2, or 4)."""
-
     @property
     def num_levels(self):
         """Number of mip levels."""
-
     @property
     def is_surface_load_store(self):
         """True if this mipmap (and each of its levels) was created with
         ``CUDA_ARRAY3D_SURFACE_LDST`` and can back a :class:`SurfaceObject`."""
-
     @property
     def device(self):
         """The :class:`Device` this mipmap was allocated on."""
+    def close(self):
+        """Release this object's reference to the underlying ``CUmipmappedArray``.
 
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc, tb):
-        ...
-
-    def __repr__(self):
-        ...
+        Destruction (``cuMipmappedArrayDestroy``) happens via the handle's
+        deleter when the last reference is dropped. A level :class:`OpaqueArray`
+        from :meth:`get_level` holds its own reference to this mipmap's storage,
+        so it stays valid until both it and this object are released. Idempotent.
+        """
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc, tb): ...
+    def __repr__(self): ...
 
 def _create_mipmapped_array(options):
     """Allocate a new :class:`MipmappedArray` on the current device.
@@ -128,4 +108,4 @@ def _create_mipmapped_array(options):
     Backs :meth:`cuda.core.Device.create_mipmapped_array`. ``options`` is a
     :class:`MipmappedArrayOptions` (or a mapping accepted by it); its fields are
     validated at construction.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/texture/_surface.pyi b/cuda_core/cuda/core/texture/_surface.pyi
index 977268abd5f..67eae47c9bb 100644
--- a/cuda_core/cuda/core/texture/_surface.pyi
+++ b/cuda_core/cuda/core/texture/_surface.pyi
@@ -1,7 +1,4 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/texture/_surface.pyx
-
-from __future__ import annotations
-
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/texture/_surface.pyx
 
 class SurfaceObject:
     """A bindless surface handle for kernel-side typed load/store.
@@ -19,38 +16,25 @@ class SurfaceObject:
 
     .. versionadded:: 1.1.0
     """
-
-    def close(self):
-        """Release this object's reference to the underlying ``CUsurfObject``.
-
-        Destruction (``cuSurfObjectDestroy``) and release of the backing array
-        happen via the handle's deleter when the last reference is dropped.
-        Idempotent.
-        """
-
-    def __init__(self, *args, **kwargs):
-        ...
-
+    def __init__(self, *args, **kwargs): ...
     @property
     def handle(self):
         """The underlying ``CUsurfObject`` as an integer (64-bit kernel arg)."""
-
     @property
     def resource(self):
         """The :class:`ResourceDescriptor` this surface was built from."""
-
     @property
-    def device(self):
-        ...
-
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc, tb):
-        ...
+    def device(self): ...
+    def close(self):
+        """Release this object's reference to the underlying ``CUsurfObject``.
 
-    def __repr__(self):
-        ...
+        Destruction (``cuSurfObjectDestroy``) and release of the backing array
+        happen via the handle's deleter when the last reference is dropped.
+        Idempotent.
+        """
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc, tb): ...
+    def __repr__(self): ...
 
 def _create_surface_object(resource):
     """Create a :class:`SurfaceObject` on the current device.
@@ -59,4 +43,4 @@ def _create_surface_object(resource):
     :class:`ResourceDescriptor` wrapping an :class:`OpaqueArray` allocated with
     ``is_surface_load_store=True``; linear/pitch2d resources are not valid
     surface backings.
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/cuda/core/texture/_texture.pyi b/cuda_core/cuda/core/texture/_texture.pyi
index 7840585bb4d..f475d8cb239 100644
--- a/cuda_core/cuda/core/texture/_texture.pyi
+++ b/cuda_core/cuda/core/texture/_texture.pyi
@@ -1,12 +1,17 @@
-# This file was generated by stubgen-pyx v0.2.6 from cuda_core/cuda/core/texture/_texture.pyx
-
-from __future__ import annotations
+# This file was generated by stubgen-pyx v0.2.19 from cuda_core/cuda/core/texture/_texture.pyx
 
 from dataclasses import dataclass
 
 from cuda.bindings import cydriver
 from cuda.core.typing import AddressModeType, FilterModeType, ReadModeType
 
+_TRSF_READ_AS_INTEGER = 1
+_TRSF_NORMALIZED_COORDINATES = 2
+_TRSF_SRGB = 16
+_TRSF_DISABLE_TRILINEAR_OPTIMIZATION = 32
+_TRSF_SEAMLESS_CUBEMAP = 64
+_ADDRESSMODE_TO_CU = {AddressModeType.WRAP: int(cydriver.CU_TR_ADDRESS_MODE_WRAP), AddressModeType.CLAMP: int(cydriver.CU_TR_ADDRESS_MODE_CLAMP), AddressModeType.MIRROR: int(cydriver.CU_TR_ADDRESS_MODE_MIRROR), AddressModeType.BORDER: int(cydriver.CU_TR_ADDRESS_MODE_BORDER)}
+_FILTERMODE_TO_CU = {FilterModeType.POINT: int(cydriver.CU_TR_FILTER_MODE_POINT), FilterModeType.LINEAR: int(cydriver.CU_TR_FILTER_MODE_LINEAR)}
 
 class ResourceDescriptor:
     """Describes the memory backing a :class:`TextureObject`.
@@ -30,13 +35,10 @@ class ResourceDescriptor:
     """
     __slots__ = ('_kind', '_source', '_format', '_num_channels', '_size_bytes', '_width', '_height', '_pitch_bytes')
 
-    def __init__(self):
-        ...
-
+    def __init__(self): ...
     @classmethod
     def from_opaque_array(cls, array):
         """Build a resource descriptor backed by a :class:`OpaqueArray`."""
-
     @classmethod
     def from_mipmapped_array(cls, mipmapped_array):
         """Build a resource descriptor backed by a :class:`MipmappedArray`.
@@ -46,7 +48,6 @@ class ResourceDescriptor:
         require a single :class:`OpaqueArray` level (obtain via
         :meth:`MipmappedArray.get_level`).
         """
-
     @classmethod
     def from_linear(cls, buffer, *, format, num_channels, size_bytes=None):
         """Build a resource descriptor for a linear (typed 1D) texture fetch.
@@ -71,7 +72,6 @@ class ResourceDescriptor:
         :class:`TextureObjectOptions` addressing/filtering fields — kernels read
         through a typed 1D fetch with bounds checking only.
         """
-
     @classmethod
     def from_pitch2d(cls, buffer, *, format, num_channels, width, height, pitch_bytes):
         """Build a resource descriptor for a row-pitched 2D image.
@@ -95,41 +95,29 @@ class ResourceDescriptor:
             ``width * format_size * num_channels`` and meet the driver's
             ``CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT``.
         """
-
     @property
-    def kind(self):
-        ...
-
+    def kind(self): ...
     @property
-    def source(self):
-        ...
-
+    def source(self): ...
     @property
     def format(self):
         """The element :class:`~cuda.core.typing.ArrayFormatType` (``None`` for array-backed)."""
-
     @property
     def num_channels(self):
         """Channels per element (``None`` for array-backed)."""
-
     @property
     def size_bytes(self):
         """Bytes bound for a linear resource (``None`` for other kinds)."""
-
     @property
     def width(self):
         """Pitch2D image width, in elements (``None`` for other kinds)."""
-
     @property
     def height(self):
         """Pitch2D image height, in rows (``None`` for other kinds)."""
-
     @property
     def pitch_bytes(self):
         """Pitch2D row pitch, in bytes (``None`` for other kinds)."""
-
-    def __repr__(self):
-        ...
+    def __repr__(self): ...
 
 @dataclass
 class TextureObjectOptions:
@@ -182,8 +170,7 @@ class TextureObjectOptions:
     max_mipmap_level_clamp: float = 0.0
     border_color: tuple[float, ...] | None = None
 
-    def __post_init__(self):
-        ...
+    def __post_init__(self): ...
 
 class TextureObject:
     """A bindless texture handle for kernel-side sampled reads.
@@ -197,61 +184,38 @@ class TextureObject:
 
     .. versionadded:: 1.1.0
     """
-
-    def close(self):
-        """Release this object's reference to the underlying ``CUtexObject``.
-
-        Destruction (``cuTexObjectDestroy``) and release of the backing resource
-        happen via the handle's deleter when the last reference is dropped.
-        Idempotent.
-        """
-
-    def __init__(self, *args, **kwargs):
-        ...
-
+    def __init__(self, *args, **kwargs): ...
     @property
     def handle(self):
         """The underlying ``CUtexObject`` as an integer (64-bit kernel arg)."""
-
     @property
     def resource(self):
         """The :class:`ResourceDescriptor` this texture was built from."""
-
     @property
     def options(self):
         """The :class:`TextureObjectOptions` this texture was built from."""
-
     @property
-    def device(self):
-        ...
-
-    def __enter__(self):
-        ...
-
-    def __exit__(self, exc_type, exc, tb):
-        ...
+    def device(self): ...
+    def close(self):
+        """Release this object's reference to the underlying ``CUtexObject``.
 
-    def __repr__(self):
-        ...
-_TRSF_READ_AS_INTEGER = 1
-_TRSF_NORMALIZED_COORDINATES = 2
-_TRSF_SRGB = 16
-_TRSF_DISABLE_TRILINEAR_OPTIMIZATION = 32
-_TRSF_SEAMLESS_CUBEMAP = 64
-_ADDRESSMODE_TO_CU = {AddressModeType.WRAP: int(cydriver.CU_TR_ADDRESS_MODE_WRAP), AddressModeType.CLAMP: int(cydriver.CU_TR_ADDRESS_MODE_CLAMP), AddressModeType.MIRROR: int(cydriver.CU_TR_ADDRESS_MODE_MIRROR), AddressModeType.BORDER: int(cydriver.CU_TR_ADDRESS_MODE_BORDER)}
-_FILTERMODE_TO_CU = {FilterModeType.POINT: int(cydriver.CU_TR_FILTER_MODE_POINT), FilterModeType.LINEAR: int(cydriver.CU_TR_FILTER_MODE_LINEAR)}
+        Destruction (``cuTexObjectDestroy``) and release of the backing resource
+        happen via the handle's deleter when the last reference is dropped.
+        Idempotent.
+        """
+    def __enter__(self): ...
+    def __exit__(self, exc_type, exc, tb): ...
+    def __repr__(self): ...
 
 def _normalize_enum(name, value, enum_type):
     """Coerce ``value`` to ``enum_type`` (a StrEnum), accepting a plain str."""
-
 def _normalize_address_modes(address_mode):
     """Return a 3-tuple of :class:`AddressModeType` values from a scalar or
     1-3 tuple. Individual entries may be plain strings."""
-
 def _create_texture_object(resource, options):
     """Create a :class:`TextureObject` on the current device.
 
     Backs :meth:`cuda.core.Device.create_texture_object`. ``resource`` is a
     :class:`ResourceDescriptor`; ``options`` is a :class:`TextureObjectOptions`
     (or a mapping accepted by it).
-    """
\ No newline at end of file
+    """
diff --git a/cuda_core/pyproject.toml b/cuda_core/pyproject.toml
index f3afa29241d..4c2c65e9b3f 100644
--- a/cuda_core/pyproject.toml
+++ b/cuda_core/pyproject.toml
@@ -155,6 +155,14 @@ ignore_missing_imports = true
 module = "cuda.core._utils.cuda_utils"
 disable_error_code = ["type-arg"]
 
+[[tool.mypy.overrides]]
+# stubgen-pyx can't resolve DLPackExchangeAPI's C function-pointer typedef
+# fields and emits bare "..." as their annotation, which is invalid outside
+# a Callable[...] context. Suppress until upstream is fixed:
+# https://github.com/jon-edward/stubgen-pyx/issues (report pending)
+module = "cuda.core._dlpack"
+disable_error_code = ["misc"]
+
 [tool.cibuildwheel]
 skip = "*-musllinux_*"
 build-verbosity = 1
diff --git a/toolshed/run_stubgen_pyx.py b/toolshed/run_stubgen_pyx.py
deleted file mode 100644
index 1a163ff0778..00000000000
--- a/toolshed/run_stubgen_pyx.py
+++ /dev/null
@@ -1,54 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-
-"""Run stubgen-pyx for cuda_core and normalize the generated stub headers.
-
-stubgen-pyx emits a path using the OS path separator in the first-line comment
-(e.g. "# This file was generated by stubgen-pyx from cuda_core\\cuda\\...").
-This wrapper rewrites that separator to "/" so committed stubs are identical
-across platforms. Line-ending normalization is handled by .gitattributes.
-
-This also forces stubgen-pyx to write files with UTF-8 encoding, which is not
-the default on Windows.
-
-This wrapper can be removed once these stubgen-pyx issues are resolved:
-    https://github.com/jon-edward/stubgen-pyx/issues/41
-    https://github.com/jon-edward/stubgen-pyx/issues/42
-"""
-
-from __future__ import annotations
-
-import os
-import pathlib
-import subprocess
-import sys
-
-_HEADER_PREFIX = b"# This file was generated by stubgen-pyx"
-
-
-def _normalize_stub_headers(root: pathlib.Path) -> None:
-    for stub in root.rglob("*.pyi"):
-        data = stub.read_bytes()
-        newline = data.find(b"\n")
-        first_line = data[:newline] if newline != -1 else data
-        if not first_line.startswith(_HEADER_PREFIX) or b"\\" not in first_line:
-            continue
-        stub.write_bytes(first_line.replace(b"\\", b"/") + data[newline:])
-
-
-def main() -> int:
-    env = os.environ.copy()
-    env.setdefault("PYTHONUTF8", "1")
-    env.setdefault("PYTHONIOENCODING", "utf-8")
-    result = subprocess.run(
-        ["stubgen-pyx", "cuda_core/cuda", "--continue-on-error", "--include-private"],  # noqa: S607
-        env=env,
-    )
-    if result.returncode != 0:
-        return result.returncode
-    _normalize_stub_headers(pathlib.Path("cuda_core/cuda"))
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())