Skip to content

Make ScalelessRMSNorm a torch.nn.RMSNorm; fix SDPACustom view -> reshape (#19376) #19376

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
navsud wants to merge 1 commit into
base: main
Choose a base branch
from
+13 −6
Open

Make ScalelessRMSNorm a torch.nn.RMSNorm; fix SDPACustom view -> reshape (#19376) #19376

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
15 changes: 11 additions & 4 deletions examples/models/llama/norm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,16 +42,23 @@ def forward(self, x):


class ScalelessRMSNorm(torch.nn.Module):
"""RMSNorm without learnable scaling.

Calls F.rms_norm with weight=None so the op composes/decomposes cleanly for
backends like QNN instead of being expressed as a hand-rolled decomposition
of mean / rsqrt / mul. Semantically equivalent to
torch.nn.RMSNorm(elementwise_affine=False), but implemented as a plain
Module to preserve the previous parameterless state_dict signature (no
`weight` attribute / parameter).
"""

def __init__(self, dim: int, eps: float = 1e-6):
super().__init__()
self.dim = dim
self.eps = eps

def forward(self, x):
x_float = x.float()
return (
x_float * torch.rsqrt((x_float * x_float).mean(-1, keepdim=True) + self.eps)
).type_as(x)
return F.rms_norm(x, (self.dim,), None, self.eps)


class RMSNormWithInputScale(torch.nn.Module):
Expand Down
4 changes: 2 additions & 2 deletions examples/models/llama/source_transformation/sdpa.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,7 @@ def forward(
0, # dropout probability. Ignored by the code
True, # is_causal
)
return output.view(bsz, seqlen, self.dim).to(dtype=input_dtype)
return output.reshape(bsz, seqlen, self.dim).to(dtype=input_dtype)


def _replace_sdpa_with_custom_op(
Expand Down Expand Up @@ -198,7 +198,7 @@ def forward(
v_scale_fp32,
)

return output.view(bsz, seqlen, self.dim)
return output.reshape(bsz, seqlen, self.dim)


def _update_attention_module_with_quantized_sdpa(
Expand Down
Loading