Skip to content

Make FactoredMatrix compatible with tensor-like arguments #599

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
JasonGross wants to merge 2 commits into
base: dev
Choose a base branch
from
Draft

Make FactoredMatrix compatible with tensor-like arguments #599

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2105225
Make `FactoredMatrix` compatible with tensor-like arguments
JasonGross May 17, 2024
7c08ab2
Merge branch 'dev' into patch-1
bryce13950 May 18, 2024
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
12 changes: 6 additions & 6 deletions transformer_lens/FactoredMatrix.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,7 +66,9 @@ def __matmul__(
"FactoredMatrix",
],
) -> Union["FactoredMatrix", Float[torch.Tensor, "... ldim"]]:
if isinstance(other, torch.Tensor):
if isinstance(other, FactoredMatrix):
return (self @ other.A) @ other.B
else:
if other.ndim < 2:
# It's a vector, so we collapse the factorisation and just return a vector
# Squeezing/Unsqueezing is to preserve broadcasting working nicely
Expand All @@ -79,8 +81,6 @@ def __matmul__(
return FactoredMatrix(self.A, self.B @ other)
else:
return FactoredMatrix(self.AB, other)
elif isinstance(other, FactoredMatrix):
return (self @ other.A) @ other.B

@overload
def __rmatmul__( # type: ignore
Expand All @@ -107,7 +107,9 @@ def __rmatmul__( # type: ignore
"FactoredMatrix",
],
) -> Union["FactoredMatrix", Float[torch.Tensor, "... rdim"]]:
if isinstance(other, torch.Tensor):
if isinstance(other, FactoredMatrix):
return other.A @ (other.B @ self)
else:
assert (
other.size(-1) == self.ldim
), f"Left matrix must match on inner dimension, shapes were self: {self.shape}, other:{other.shape}"
Expand All @@ -118,8 +120,6 @@ def __rmatmul__( # type: ignore
return FactoredMatrix(other @ self.A, self.B)
else:
return FactoredMatrix(other, self.AB)
elif isinstance(other, FactoredMatrix):
return other.A @ (other.B @ self)

def __mul__(self, scalar: Union[int, float, torch.Tensor]) -> FactoredMatrix:
"""
Expand Down