mirror of
https://github.com/pytorch/pytorch.git
synced 2025-11-11 22:34:53 +08:00
[ci] disable cudagraph for tts_angular on dashboard (#148221)
tts_angular with cudagraph is flaky. Its speedup varies from .05 to 1.01. This PR disables cudagraph for tts_angular to avoid the noise. Since tts_angular shows ~1x speedup while other torchbench models show ~2x speedup, skipping tts_angular would wrongly bump the cudagraph speedup. So this PR only disables cudagraph for tts_angular instead of skipping tts_angular. [Dashboard ](https://github.com/pytorch/pytorch/actions/runs/13597394087) Pull Request resolved: https://github.com/pytorch/pytorch/pull/148221 Approved by: https://github.com/eellison
This commit is contained in:
committed by
PyTorch MergeBot
parent
de7af81f18
commit
6e10471966
@ -1824,6 +1824,10 @@ class BenchmarkRunner:
|
||||
def skip_models_due_to_control_flow(self):
|
||||
return set()
|
||||
|
||||
@property
|
||||
def disable_cudagraph_models(self):
|
||||
return set()
|
||||
|
||||
@property
|
||||
def guard_on_nn_module_models(self):
|
||||
return set()
|
||||
@ -3834,6 +3838,9 @@ def run(runner, args, original_dir=None):
|
||||
experiment = coverage_experiment
|
||||
output_filename = "coverage.csv"
|
||||
|
||||
if args.only in runner.disable_cudagraph_models:
|
||||
args.disable_cudagraphs = True
|
||||
|
||||
if args.inductor or args.backend == "inductor" or args.export_aot_inductor:
|
||||
inductor_config.triton.cudagraphs = not args.disable_cudagraphs
|
||||
inductor_config.triton.persistent_reductions = (
|
||||
|
||||
@ -145,6 +145,10 @@ class TorchBenchmarkRunner(BenchmarkRunner):
|
||||
def skip_models_for_freezing_cuda(self):
|
||||
return self._skip["freezing"]["cuda"]
|
||||
|
||||
@property
|
||||
def disable_cudagraph_models(self):
|
||||
return self._config["disable_cudagraph"]
|
||||
|
||||
@property
|
||||
def skip_models_for_freezing_cpu(self):
|
||||
return self._skip["freezing"]["cpu"]
|
||||
|
||||
@ -111,6 +111,12 @@ non_deterministic:
|
||||
- sam_fast
|
||||
|
||||
|
||||
disable_cudagraph:
|
||||
# tts_angular is flaky with cudagraphs. Its speedup
|
||||
# oscillates from .05 to 1.05
|
||||
- tts_angular
|
||||
|
||||
|
||||
dtype:
|
||||
force_amp_for_fp16_bf16_models:
|
||||
- DALLE2_pytorch
|
||||
|
||||
Reference in New Issue
Block a user