[Hardware][GPU] Profiler config additional to increase it scope and annotation details (#37524)

Signed-off-by: devalshahamd <deval.shah@amd.com>
Signed-off-by: Deval Shah <devashah@amd.com>
Signed-off-by: Deval Shah <deval.shah@amd.com>
Co-authored-by: Deval Shah <devashah@amd.com>
This commit is contained in:
devalshahamd
2026-07-17 13:38:59 -07:00
committed by GitHub
co-authored by Deval Shah
parent b5433b6f50
commit fcd2255d16
4 changed files with 240 additions and 26 deletions
+17
View File
@@ -66,6 +66,17 @@ class ProfilerConfig:
"""If `True`, enables memory profiling in the torch profiler.
Disabled by default."""
capture_torch_profiler: bool = False
"""If `True`, enables a torch profiler during CUDA graph capture on rank 0.
Traces are saved to a `capture_traces` subdirectory under `torch_profiler_dir`.
Requires `profiler` to be set to 'torch'."""
detailed_trace_annotation: bool = False
"""If `True`, uses detailed annotations with roofline metrics (sk, sqsq,
sqsk) in profiler trace events. If `False`, uses simple annotations with
only context/generation request counts and token counts.
Disabled by default."""
ignore_frontend: bool = False
"""If `True`, disables the front-end profiling of AsyncLLM when using the
'torch' profiler. This is needed to reduce overhead when using delay/limit options,
@@ -144,4 +155,10 @@ class ProfilerConfig:
if profiler_dir and not _is_uri_path(profiler_dir):
self.torch_profiler_dir = os.path.abspath(os.path.expanduser(profiler_dir))
if self.capture_torch_profiler and self.profiler != "torch":
raise ValueError(
"capture_torch_profiler is only applicable when profiler is "
"set to 'torch'"
)
return self