torch.mps

Created On: Feb 10, 2023 | Last Updated On: Jun 08, 2025

This package enables an interface for accessing MPS (Metal Performance Shaders) backend in Python. Metal is Apple’s API for programming metal GPU (graphics processor unit). Using MPS means that increased performance can be achieved, by running work on the metal GPU(s). See https://developer.apple.com/documentation/metalperformanceshaders for more details.

device_count	Returns the number of available MPS devices.
synchronize	Waits for all kernels in all streams on a MPS device to complete.
get_rng_state	Returns the random number generator state as a ByteTensor.
set_rng_state	Sets the random number generator state.
manual_seed	Sets the seed for generating random numbers.
seed	Sets the seed for generating random numbers to a random number.
empty_cache	Releases all unoccupied cached memory currently held by the caching allocator so that those can be used in other GPU applications.
set_per_process_memory_fraction	Set memory fraction for limiting process's memory allocation on MPS device.
current_allocated_memory	Returns the current GPU memory occupied by tensors in bytes.
driver_allocated_memory	Returns total GPU memory allocated by Metal driver for the process in bytes.
recommended_max_memory	Returns recommended max Working set size for GPU memory in bytes.
compile_shader	Compiles compute shader from source and allows one to invoke kernels defined there from the comfort of Python runtime Example.

MPS Profiler

profiler.start	Start OS Signpost tracing from MPS backend.
profiler.stop	Stops generating OS Signpost tracing from MPS backend.
profiler.profile	Context Manager to enabling generating OS Signpost tracing from MPS backend.
profiler.is_capturing_metal	Cheks if metal capture is in progress
profiler.is_metal_capture_enabled	Checks if metal_capture context manager is usable To enable metal capture, set MTL_CAPTURE_ENABLED envvar
profiler.metal_capture	Conext manager that enables capturing of Metal calls into gputrace

MPS Event

event.Event

Wrapper around an MPS event.