Skip to content

[TPU][Pallas] relax tolerances and fix Pallas autotuning OOM in layer_norm #2272

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
yarongmu-google wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

[TPU][Pallas] relax tolerances and fix Pallas autotuning OOM in layer_norm #2272

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
48 changes: 42 additions & 6 deletions examples/layer_norm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,20 @@


# %%
@helion.kernel
def baseline_ln_fwd(
x: torch.Tensor,
normalized_shape: list[int],
weight: torch.Tensor,
bias: torch.Tensor | None = None,
eps: float = 1e-5,
) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
mean = x.to(torch.float32).mean(dim=-1)
var = x.to(torch.float32).var(dim=-1, unbiased=False)
rstd = torch.rsqrt(var + eps)
out = torch.nn.functional.layer_norm(x, normalized_shape, weight, bias, eps)
return out, mean, rstd

@helion.kernel(autotune_baseline_fn=baseline_ln_fwd)
def layer_norm_fwd(
x: torch.Tensor,
normalized_shape: list[int],
Expand Down Expand Up @@ -83,7 +96,30 @@ def layer_norm_fwd(


# %%
@helion.kernel
def baseline_ln_bwd(
grad_out: torch.Tensor,
x: torch.Tensor,
mean: torch.Tensor,
rstd: torch.Tensor,
weight: torch.Tensor,
compute_bias_grad: bool = True,
) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor | None]:
x_hat = (x - mean[:, None]) * rstd[:, None]
grad_weight = (grad_out * x_hat).sum(dim=0).to(weight.dtype)
grad_bias = grad_out.sum(dim=0).to(weight.dtype) if compute_bias_grad else None

n = x.size(1)
wdy = weight * grad_out
c1 = (x_hat * wdy).sum(dim=-1, keepdim=True) / n
c2 = wdy.sum(dim=-1, keepdim=True) / n
grad_x = ((wdy - (x_hat * c1 + c2)) * rstd[:, None]).to(x.dtype)

return grad_x, grad_weight, grad_bias

@helion.kernel(
autotune_baseline_fn=baseline_ln_bwd,
config=helion.Config(block_sizes=[32, 1024]),
)
def layer_norm_bwd(
grad_out: torch.Tensor,
x: torch.Tensor,
Expand Down Expand Up @@ -259,8 +295,8 @@ def main() -> None:
layer_norm,
torch.nn.functional.layer_norm,
(x, [dim], weight, b, eps),
rtol=1e-3,
atol=1e-3,
rtol=1e-2,
atol=1e-2,
)

# Test forward + backward pass
Expand All @@ -277,8 +313,8 @@ def main() -> None:
layer_norm,
torch.nn.functional.layer_norm,
(x_grad, [dim], weight_grad, b, eps),
rtol=1e-3,
atol=1e-3,
rtol=1e-2,
atol=1e-2,
bwd=True,
)

Expand Down
Loading