Skip to content

perf(dex-hand): run retargeting optimizer under torch.no_grad #478

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

Open
nvddr wants to merge 2 commits into
base: main
Choose a base branch
from
Open

perf(dex-hand): run retargeting optimizer under torch.no_grad #478

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c176799
perf(dex-hand): run retargeting optimizer under torch.no_grad
nvddr May 7, 2026
8b399e1
review(dex-hand): preserve inference_mode(False) escape
nvddr May 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
8 changes: 7 additions & 1 deletion src/retargeters/dex_hand_retargeter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -412,10 +412,16 @@ def _compute_hand(self, poses: Dict[str, np.ndarray]) -> np.ndarray:
ref_value = target_pos[task_indices, :] - target_pos[origin_indices, :]

# 4. Run optimizer
# ``dex_retargeting`` solves a QP-style optimization that does not
# require autograd. ``torch.no_grad()`` avoids the per-step
# grad-tracking overhead the previous ``enable_grad`` context paid for.
# ``torch.inference_mode(False)`` is preserved so callers running
# inside an outer ``torch.inference_mode()`` (where some in-place /
# view ops can error) still see the optimizer execute in normal mode.
try:
import torch # type: ignore

with torch.enable_grad(), torch.inference_mode(False):
with torch.inference_mode(False), torch.no_grad():
return self._dex_hand.retarget(ref_value) # type: ignore
Comment on lines +415 to 425
except Exception as e:
logger.error(f"Error in retargeting: {e}")
Expand Down
Loading