Skip to content
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

Specify output dtype to torch.float32 in _foreach_norm #727

Merged
merged 1 commit into from
Aug 22, 2024
Merged

Specify output dtype to torch.float32 in _foreach_norm #727

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
4 changes: 2 additions & 2 deletions torchao/float8/fsdp_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,15 +59,15 @@ def precompute_float8_dynamic_scale_for_fsdp(module: nn.Module) -> None:
return

# inf-norm is equivalent to max(abs(w))
max_weights = torch._foreach_norm(weights, ord=math.inf) # Partial
max_weights = torch._foreach_norm(weights, ord=math.inf, dtype=torch.float32) # Partial
amax_tensor = torch.stack(max_weights) # Partial
# clamp is dispatched through DTensor
# it will issue a single all-reduce
amax_tensor = torch.clamp(amax_tensor, EPS) # Replicate
scale_tensor = torch.finfo(torch.float8_e4m3fn).max / amax_tensor # Replicate
if amax_tensor.dtype is torch.float16:
scale_tensor = torch.clamp(scale_tensor, max=torch.finfo(torch.float16).max)
local_scale_tensor = scale_tensor.to_local().to(dtype=torch.float32)
local_scale_tensor = scale_tensor.to_local()
for i, float8_linear in enumerate(float8_linears):
float8_linear.weight._local_tensor._precomputed_scale = local_scale_tensor[i]

Expand Down
Loading