facebookresearch/HolisticTraceAnalysis

Support the Equivalent of profiler.key_averages(group_by_stack_n=n)

Aperta

#294 aperta il 3 set 2025

 (1 commento) (0 reazioni) (0 assegnatari)Python (94 fork)auto 404
help wantedneeds triagequestion

Metriche repository

Star
 (535 stelle)
Metriche merge PR
 (Metriche PR in attesa)

Descrizione

What is your question?

The pytorch profiler can aggregate runtimes by the operation and its traceback via key_averages(group_by_stack_n=n): https://docs.pytorch.org/tutorials/beginner/profiler.html#print-profiler-result

Identifying the top operators / kernels and their source user Python code is helpful to find the main bottlenecks and perform targeted optimizations. Does HTA support something equivalent, or could you shed some light on implementing one? Thanks!

I also found get_aten_op_kernels_and_delay(), but seems like it traces down from aten ops only instead of source user training code.

Code

No response

What have you tried?

No response

Environment

No response

Guida contributor