Skip to content

perf: zero_grad(set_to_none=True) and reduce checkpoint I/O #83

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
anxiangsir merged 2 commits into from
Feb 10, 2026
Merged

perf: zero_grad(set_to_none=True) and reduce checkpoint I/O #83

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
41b64cd
perf: use zero_grad(set_to_none=True) to reduce memory bandwidth
Luodian Feb 7, 2026
e2be8ff
perf: only save HuggingFace checkpoint at final step to reduce I/O st…
Luodian Feb 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
4 changes: 1 addition & 3 deletions training/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -664,7 +664,7 @@ def wrap_ddp(model):
for pfc in list_module_pfc:
clip_grad_norm_(pfc.parameters(), max_norm=5, norm_type=2)
opt.step()
opt.zero_grad()
opt.zero_grad(set_to_none=True)

lr_scheduler.step()

Expand Down Expand Up @@ -692,8 +692,6 @@ def wrap_ddp(model):
list_head_names=args.list_head_names,
keep_num=20,
)
# Also save in HuggingFace format
save_hf_checkpoint(args.output, backbone, global_step=global_step, image_size=args.image_size[0])

if global_step > args.total_steps:
save_checkpoint(
Expand Down