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.
An earlier version of the API in
torch.autograd module is considered legacy and will be deprecated.
profile(*, activities=None, schedule=None, on_trace_ready=None, record_shapes=False, profile_memory=False, with_stack=False, use_gpu=None)¶
Profiler context manager.
activities- list of activity groups (CPU, CUDA) to use in profiling;
schedule- callable that takes step (int) as a single parameter and returns
ProfilerActionvalue 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
torch.profiler.scheduleto 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.
Enabling shape and stack tracing results in additional overhead.
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
# 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_num) + ".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.step()
Exports the collected trace in Chrome JSON format.
Save stack traces in a file in a format suitable for visualization.
path- save stacks file to this location;
metric- metric to use: “self_cpu_time_total” or “self_cuda_time_total”
Example of using FlameGraph tool:
./flamegraph.pl –title “CPU time” –countname “us.” profiler.stacks > perf_viz.svg
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.
Signals the profiler that the next profiling step has started.
schedule(*, wait, warmup, active)¶
Represents profiler behavior: wait for
waitsteps, then do the warmup for the next
warmupsteps, then do the active recording for the next
activesteps and then repeat the cycle staring with the next step.