Skip to content

docs: Fix incorrect column-major scale layout in FP8 GEMM docstrings #2614

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
bledden wants to merge 2 commits into
base: main
Choose a base branch
from
Open

docs: Fix incorrect column-major scale layout in FP8 GEMM docstrings #2614

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4410a61
docs: Fix incorrect column-major scale layout description in FP8 GEMM…
bledden Feb 21, 2026
2b84047
fix: also fix adjacent scale_major_k typo in b_scale docstring
bledden Feb 22, 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
8 changes: 4 additions & 4 deletions flashinfer/gemm/gemm_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4203,15 +4203,15 @@ def gemm_fp8_nt_groupwise(

a_scale: torch.Tensor
if the backend is ``cutlass``:
Column-major scale tensor for a, shape ``(m, k // block_size)`` if scale_major_mode is ``K``
Row-major scale tensor for a, shape ``(m, k // block_size)`` if scale_major_mode is ``K``
or shape ``(k // block_size, m)`` if scale_major_mode is ``MN``
if the backend is ``trtllm``:
scale_major_mode should be None, the scale tensor should be (m, k // block_size),
contiguous on the first dimension

b_scale: torch.Tensor
if the backend is ``cutlass``:
Row-major scale tensor for b, shape ``(n // block_size, k // block_size)`` if scale_major_k is ``K``
Row-major scale tensor for b, shape ``(n // block_size, k // block_size)`` if scale_major_mode is ``K``
or shape ``(k // block_size, n // block_size)`` if scale_major_mode is ``MN``
if the backend is ``trtllm``:
scale_major_mode should be None, the scale tensor should be (k // block_size, n // block_size),
Expand Down Expand Up @@ -4573,7 +4573,7 @@ def group_gemm_fp8_nt_groupwise(
Column-major input tensor shape ``(batch_size, n, k)``, data type is ``torch.float8_e4m3fn`` or ``torch.float8_e5m2``.

a_scale: torch.Tensor
Column-major scale tensor for a, shape ``(cum_m, k // block_size)`` if scale_major_mode is ``K``
Row-major scale tensor for a, shape ``(cum_m, k // block_size)`` if scale_major_mode is ``K``
or shape ``(k // block_size, cum_m)`` if scale_major_mode is ``MN``, data type is ``torch.float32``.

b_scale: torch.Tensor
Expand Down Expand Up @@ -4783,7 +4783,7 @@ def group_gemm_mxfp8_mxfp4_nt_groupwise(
Column-major input tensor, shape ``(batch_size, n, k // 2)``, data type is ``torch.uint8``.

a_scale: torch.Tensor
Column-major scale tensor for a, shape ``(cum_m_padded, k // 32)``, data type is ``torch.uint8``.
Row-major scale tensor for a, shape ``(cum_m_padded, k // 32)``, data type is ``torch.uint8``.

b_scale: torch.Tensor
Row-major scale tensor for b, shape ``(batch_size, n_padded, k // 32)``, data type is ``torch.uint8``.
Expand Down