Skip to content

perf(autogram): Avoid tensor creation in autogram hook #820

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

Merged
ValerianRey merged 2 commits into from
Oct 7, 2026
Merged

perf(autogram): Avoid tensor creation in autogram hook #820

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
910c2f2
Avoid tensor creation in autogram hook
ValerianRey Oct 7, 2026
ac2a771
Merge branch 'main' into perf/autogram-hook-smallest-output
ValerianRey Oct 7, 2026
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
7 changes: 3 additions & 4 deletions src/torchjd/autogram/_module_hook_manager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
import weakref
from typing import Any, cast
from typing import Any

import torch
from torch import Tensor, nn
Expand Down Expand Up @@ -129,9 +129,8 @@ def __call__(
# We only care about running the AutogramNode, so we need one of its child
# edges (the edges of the original outputs of the model) as target. For memory
# efficiency, we select the smallest one (that requires grad).
preference = torch.tensor([t.numel() for t in rg_outputs])
index = cast(int, preference.argmin().item())
self.target_edges.register(get_gradient_edge(rg_outputs[index]))
smallest_rg_output = min(rg_outputs, key=lambda t: t.numel())
self.target_edges.register(get_gradient_edge(smallest_rg_output))

autograd_fn_rg_outputs = AutogramNode.apply(
self.gramian_accumulation_phase,
Expand Down
Loading