Skip to content

Disable AMP for preconditioner calculations #833

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
AnnaKwa wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Disable AMP for preconditioner calculations #833

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d9449c1
Keep preconditioning coefs in float32, use amp for unet call
AnnaKwa Feb 12, 2026
39e10ac
disable autocast for most of precond forward
AnnaKwa Feb 12, 2026
7c98f43
Merge branch 'fix/move-amp-after-precond' of github.com:ai2cm/ace int…
AnnaKwa Feb 12, 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
48 changes: 25 additions & 23 deletions fme/downscaling/modules/preconditioners.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,30 +86,31 @@ def forward(
class_labels=None,
force_fp32=False,
):
x = x.to(torch.float32)
sigma = sigma.to(torch.float32).reshape(-1, 1, 1, 1)
class_labels = (
None
if self.label_dim == 0
else torch.zeros([1, self.label_dim], device=x.device)
if class_labels is None
else class_labels.to(torch.float32).reshape(-1, self.label_dim)
)
dtype = (
torch.float16
if (self.use_fp16 and not force_fp32 and x.device.type == "cuda")
else torch.float32
)
with torch.amp.autocast(device_type="cuda", enabled=False):
x = x.to(torch.float32)
sigma = sigma.to(torch.float32).reshape(-1, 1, 1, 1)
class_labels = (
None
if self.label_dim == 0
else torch.zeros([1, self.label_dim], device=x.device)
if class_labels is None
else class_labels.to(torch.float32).reshape(-1, self.label_dim)
)
dtype = (
torch.float16
if (self.use_fp16 and not force_fp32 and x.device.type == "cuda")
else torch.float32
)

c_skip = self.sigma_data**2 / (sigma**2 + self.sigma_data**2)
c_out = sigma * self.sigma_data / (sigma**2 + self.sigma_data**2).sqrt()
c_in = 1 / (self.sigma_data**2 + sigma**2).sqrt()
c_noise = sigma.log() / 4
c_skip = self.sigma_data**2 / (sigma**2 + self.sigma_data**2)
c_out = sigma * self.sigma_data / (sigma**2 + self.sigma_data**2).sqrt()
c_in = 1 / (self.sigma_data**2 + sigma**2).sqrt()
c_noise = sigma.log() / 4

arg = c_in * x
arg = c_in * x

if condition is not None:
arg = torch.cat([arg, condition], dim=1)
if condition is not None:
arg = torch.cat([arg, condition], dim=1)

F_x = self.model(
arg.to(dtype),
Expand All @@ -121,5 +122,6 @@ def forward(
raise ValueError(
f"Expected the dtype to be {dtype}, but got {F_x.dtype} instead."
)
D_x = c_skip * x + c_out * F_x.to(torch.float32)
return D_x
with torch.amp.autocast(device_type=get_device().type, enabled=False):
D_x = c_skip * x + c_out * F_x.to(torch.float32)
return D_x