Skip to content

fix abnormal spike : Init res state to 1.0 #6

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
Maru-mee wants to merge 2 commits into
base: master
Choose a base branch
from
Open

fix abnormal spike : Init res state to 1.0 #6

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
28458db
Init res state to 1.0
Maru-mee Sep 29, 2025
362737e
r_factor preventing division by zero
Maru-mee Sep 29, 2025
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: 4 additions & 4 deletions came_pytorch/CAME.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,7 @@ def _rms(self, tensor):

def _approx_sq_grad(self, exp_avg_sq_row, exp_avg_sq_col):
r_factor = (
(exp_avg_sq_row / exp_avg_sq_row.mean(dim=-1, keepdim=True))
(exp_avg_sq_row / exp_avg_sq_row.mean(dim=-1, keepdim=True).add_(self.param_groups[0]["eps"][1]))
.rsqrt_()
.unsqueeze(-1)
)
Expand Down Expand Up @@ -102,8 +102,8 @@ def step(self, closure=None):
grad_shape[:-2] + grad_shape[-1:]
).type_as(grad)

state["exp_avg_res_row"] = torch.zeros(grad_shape[:-1]).type_as(grad)
state["exp_avg_res_col"] = torch.zeros(
state["exp_avg_res_row"] = torch.ones(grad_shape[:-1]).type_as(grad)
state["exp_avg_res_col"] = torch.ones(
grad_shape[:-2] + grad_shape[-1:]
).type_as(grad)
else:
Expand Down Expand Up @@ -171,4 +171,4 @@ def step(self, closure=None):
update.mul_(group["lr"])
p.data.add_(-update)

return loss
return loss