Skip to content
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

Non-reentrant checkpointing hook fix #5781

Merged
merged 7 commits into from
Aug 2, 2024
Merged

Non-reentrant checkpointing hook fix #5781

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
98972b4
added condition to non-reentrant checkpointing fn
ic-synth Jul 18, 2024
1479cbf
Merge branch 'master' into checkpoint-hook-fix
HeyangQin Jul 23, 2024
66453a3
Merge branch 'master' into checkpoint-hook-fix
loadams Aug 1, 2024
f9f63e2
Merge branch 'master' into checkpoint-hook-fix
loadams Aug 1, 2024
dd1bfb3
Merge branch 'master' into checkpoint-hook-fix
loadams Aug 1, 2024
9565750
Merge branch 'master' into checkpoint-hook-fix
loadams Aug 1, 2024
4ecd9e3
Merge branch 'master' into checkpoint-hook-fix
loadams Aug 1, 2024
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
5 changes: 3 additions & 2 deletions deepspeed/runtime/activation_checkpointing/checkpointing.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -965,8 +965,9 @@ def after_backward_hook(_nonuse_grads):

with torch.autograd.graph.saved_tensors_hooks(checkpoint_pack, checkpoint_unpack):
outputs = function(*inputs_cuda)
for leaf_tensor in leaf_tensors:
leaf_tensor.register_hook(after_backward_hook)
if PROFILE_TIME or SYNCHRONIZE:
for leaf_tensor in leaf_tensors:
leaf_tensor.register_hook(after_backward_hook)

see_memory_usage("After running forward on the layer", force=False)

Expand Down
Loading