Rate this Page
★★★★★

record_function#

classtorch.autograd.profiler.record_function(name,args=None)[source]#

Context manager/function decorator that adds a label to a code block/function when running autograd profiler.Label will only appear if CPU activity tracing is enabled.

It is useful when tracing the code profile.

Parameters

  • name (str) – Label assigned to the block of code.

  • node_id (int) – ID of node, for distributed profiling. Unset in

  • cases. (non-distributed) –

Example

>>>x=torch.randn((1,1),requires_grad=True)>>>withtorch.autograd.profiler.profile()asprof:...y=x**2...withtorch.autograd.profiler.record_function(..."label-z"...):# label the block...z=y**3...y.backward()>>># NOTE: some columns were removed for brevity>>>print(prof.key_averages().table(sort_by="self_cpu_time_total"))-----------------------------------  ---------------  ---------------  ---------------Name                                 Self CPU total %  CPU time avg     Number of Calls-----------------------------------  ---------------  ---------------  ---------------pow                                  60.77%           47.470us         3mul                                  21.73%           25.465us         2PowBackward0                         12.03%           121.891us        1torch::autograd::AccumulateGrad      2.70%            6.324us          1label-z                              2.13%            12.421us         1torch::autograd::GraphRoot           0.64%            1.503us          1-----------------------------------  ---------------  ---------------  ---------------Self CPU time total: 234.344usCUDA time total: 0.000us