Skip to content

Fix hybrid extra state size mismatch #3499

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

Draft
negvet wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Fix hybrid extra state size mismatch #3499

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
14 changes: 11 additions & 3 deletions tests/pytorch/distributed/fsdp2_tests/run_fsdp2_fused_adam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2192,8 +2192,13 @@ def run_continuation_step(current_model, current_optimizer):
with te.autocast(enabled=True, recipe=hybrid_recipe):
ref_output = model(x).clone()

# CustomRecipe's _extra_state pickle size isn't guaranteed across models/
# ranks, so strip it like the DelayedScaling case above (#1860).
model_state = {
k: v for k, v in model.state_dict().items() if not k.endswith("_extra_state")
}
save_state = {
"model": model.state_dict(),
"model": model_state,
"optimizer": optimizer.state_dict(),
}
dcp.save(save_state, checkpoint_id=checkpoint_dir)
Expand All @@ -2213,12 +2218,15 @@ def run_continuation_step(current_model, current_optimizer):
F.mse_loss(out_tmp, target).backward()
optimizer2.step()

model2_state = {
k: v for k, v in model2.state_dict().items() if not k.endswith("_extra_state")
}
state_to_load = {
"model": model2.state_dict(),
"model": model2_state,
"optimizer": optimizer2.state_dict(),
}
dcp.load(state_to_load, checkpoint_id=checkpoint_dir)
model2.load_state_dict(state_to_load["model"])
model2.load_state_dict(state_to_load["model"], strict=False)
optimizer2.load_state_dict(state_to_load["optimizer"])

with torch.no_grad():
Expand Down
Loading