Skip to content

[Profiler] Memory profiler part 10: Mark optimizer state #88925

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
robieta wants to merge 4 commits into from
Closed

[Profiler] Memory profiler part 10: Mark optimizer state #88925

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
db78517
[Profiler] Memory profiler part 10: Mark optimizer state
Nov 12, 2022
7712009
Update on "[Profiler] Memory profiler part 10: Mark optimizer state"
Nov 19, 2022
cf0f7e3
Update on "[Profiler] Memory profiler part 10: Mark optimizer state"
Nov 22, 2022
876be4f
Update on "[Profiler] Memory profiler part 10: Mark optimizer state"
Nov 26, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 8 additions & 8 deletions test/profiler/test_memory_profiler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1305,14 +1305,14 @@ def step_fn(mark_region):
aten::detach 7 (GRADIENT) -> 7 (GRADIENT)

-- Optimizer --------------------------------------------------------------------------------------------
aten::clone 7 (GRADIENT) -> 10 (???)
aten::detach 10 (???) -> 10 (???)
aten::detach 10 (???) -> 10 (???)
aten::add_.Tensor 2 (PARAMETER), 10 (???) -> 2 (PARAMETER)
aten::clone 9 (GRADIENT) -> 11 (???)
aten::detach 11 (???) -> 11 (???)
aten::detach 11 (???) -> 11 (???)
aten::add_.Tensor 3 (PARAMETER), 11 (???) -> 3 (PARAMETER)
aten::clone 7 (GRADIENT) -> 10 (OPTIMIZER_STATE)
aten::detach 10 (OPTIMIZER_STATE) -> 10 (OPTIMIZER_STATE)
aten::detach 10 (OPTIMIZER_STATE) -> 10 (OPTIMIZER_STATE)
aten::add_.Tensor 2 (PARAMETER), 10 (OPTIMIZER_STATE) -> 2 (PARAMETER)
aten::clone 9 (GRADIENT) -> 11 (OPTIMIZER_STATE)
aten::detach 11 (OPTIMIZER_STATE) -> 11 (OPTIMIZER_STATE)
aten::detach 11 (OPTIMIZER_STATE) -> 11 (OPTIMIZER_STATE)
aten::add_.Tensor 3 (PARAMETER), 11 (OPTIMIZER_STATE) -> 3 (PARAMETER)
aten::zero_ 7 (GRADIENT) -> 7 (GRADIENT)
aten::zero_ 9 (GRADIENT) -> 9 (GRADIENT)""",
)
Expand Down
12 changes: 12 additions & 0 deletions torch/profiler/_memory_profiler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
import collections
import dataclasses
import enum
import itertools as it
from typing import (
Any,
cast,
Expand Down Expand Up @@ -36,6 +37,7 @@ class Category(enum.Enum):
ACTIVATION = enum.auto()
GRADIENT = enum.auto()
PARAMETER = enum.auto()
OPTIMIZER_STATE = enum.auto()


@dataclasses.dataclass
Expand Down Expand Up @@ -561,6 +563,7 @@ def __init__(self, result: _ProfilerResult) -> None:
self._set_inputs()
self._set_parameters_using_data_flow()
self._set_activations()
self._set_optimizer_state()

def _is_gradient(self, *args, **kwargs) -> bool:
return self._categories.get(*args, **kwargs) == Category.GRADIENT
Expand Down Expand Up @@ -781,3 +784,12 @@ def _set_activations(self) -> None:
):
for i in node.outputs.items():
self._categories.setdefault_by_version(*i, Category.ACTIVATION)

def _set_optimizer_state(self) -> None:
for event in self._op_tree.dfs():
if event.typed[0] == _EventType.PyCall and event.typed[1].optimizer:
parameters = event.typed[1].optimizer.parameters
for _, t in it.chain(*[state for _, _, state in parameters]):
key = TensorKey.from_tensor(t)
if key is not None:
self._categories.set_by_id(key, Category.OPTIMIZER_STATE)