Skip to content

[CPU offload optim] Fix when there are non-trainable params #1210

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
merged 1 commit into from
Nov 2, 2024
Merged

[CPU offload optim] Fix when there are non-trainable params #1210

Show file tree
Hide file tree
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
1 change: 1 addition & 0 deletions test/prototype/test_low_bit_optim.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,6 +211,7 @@ def test_optim_4bit_correctness(self, optim_name):
def test_optim_cpu_offload_correctness(self, offload_grad, grad_accum):
device = "cuda"
model1 = nn.Sequential(nn.Linear(32, 1024), nn.ReLU(), nn.Linear(1024, 128)).to(device)
model1[0].requires_grad_(False) # make sure it can work in the presence of non-trainable params
model2 = copy.deepcopy(model1)

optim1 = torch.optim.AdamW(model1.parameters())
Expand Down
3 changes: 3 additions & 0 deletions torchao/prototype/low_bit_optim/cpu_offload.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,9 @@ def backward_hook(p_cuda):
params = param_group.pop("params")

for p_cuda in params:
if not p_cuda.requires_grad:
continue

# pre-allocate CPU params and grads
p_cpu = torch.empty_like(p_cuda, device="cpu", pin_memory=True)
p_cpu.grad = torch.empty_like(p_cpu, pin_memory=True)
Expand Down
Loading