torch.profiler

Overview

PyTorch Profiler is a tool that allows the collecton of the performance metrics during the training and inference. Profiler’s context manager API can be used to better understand what model operators are the most expensive, examine their input shapes and stack traces, study device kernel activity and visualize the execution trace.

Note

An earlier version of the API in torch.autograd module is considered legacy and will be deprecated.

API Reference

class torch.profiler.profile(*, activities=None, schedule=None, on_trace_ready=None, record_shapes=False, profile_memory=False, with_stack=False, use_gpu=None)

Profiler context manager.

Arguments:

• activities - list of activity groups (CPU, CUDA) to use in profiling;

• schedule - callable that takes step (int) as a single parameter and returns ProfilerAction value that specifies the profiler action on each step;

• on_trace_ready (optional) - callable, called each time the trace is ready during the profiling;

• record_shapes - save information about operator’s input shapes;

• profile_memory - track tensor memory allocation/deallocation;

• with_stack - save stack traces;

• use_gpu - (deprecated, use activities).

Note

Use torch.profiler.schedule to generate the callable schedule. Non-default schedules are useful when profiling long training jobs and allow the user to obtain multiple traces at the different iterations of the training process. The default schedule simply records all the events continuously for the duration of the context manager.

Note

Enabling shape and stack tracing results in additional overhead.

Examples:

with torch.profiler.profile(
    activities=[
        torch.profiler.ProfilerActivity.CPU,
        torch.profiler.ProfilerActivity.CUDA]
) as p:
    code_to_profile()
print(p.key_averages().table(
    sort_by="self_cuda_time_total", row_limit=-1))

Usimg the profiler’s schedule, on_trace_ready and next_step functions:

# Non-default profiler schedule allows user to turn profiler on and off
# on different iterations of the training loop;
# trace_handler is called every time a new trace becomes available
def trace_handler(prof):
    print(prof.key_averages().table(
        sort_by="self_cuda_time_total", row_limit=-1))
    # prof.export_chrome_trace("/tmp/test_trace_" + str(prof.step()) + ".json")

with torch.profiler.profile(
    activities=[
        torch.profiler.ProfilerActivity.CPU,
        torch.profiler.ProfilerActivity.CUDA],

    # In this example with wait=1, warmup=1, active=2,
    # profiler will skip the first step/iteration,
    # start warming up on the second, record
    # the third and the forth iterations,
    # after which the trace will become available
    # and on_trace_ready (when set) is called;
    # the cycle repeats starting with the next step

    schedule=torch.profiler.schedule(
        wait=1,
        warmup=1,
        active=2),
    on_trace_ready=trace_handler
    ) as p:
        for iter in range(N):
            code_iteration_to_profile(iter)
            # send a signal to the profiler that the next iteration has started
            p.next_step()

export_chrome_trace(path)

Exports the collected trace in Chrome JSON format.

export_stacks(path, metric='self_cpu_time_total')

Save stack traces in a file in a format suitable for visualization.

Arguments:

• path - save stacks file to this location;

• metric - metric to use: “self_cpu_time_total” or “self_cuda_time_total”

Note

Example of using FlameGraph tool:

• git clone https://github.com/brendangregg/FlameGraph

• cd FlameGraph

• ./flamegraph.pl –title “CPU time” –countname “us.” profiler.stacks > perf_viz.svg

key_averages(group_by_input_shape=False, group_by_stack_n=0)

Averages events, grouping them by operator name and (optionally) input shapes and stack. Note: to use shape/stack functionality make sure to set record_shapes/with_stack when creating profiler context manager.

next_step()

Signals the profiler that the next profiling step has started.

step()

Returns the current profiling step.

torch.profiler.schedule(*, wait, warmup, active)

Represents profiler behavior: wait for wait steps, then do the warmup for the next warmup steps, then do the active recording for the next active steps and then repeat the cycle staring with the next step.