[Bugfix] Fix NVFP4 fused SiLU+mul scale allocation and global scale direction (#55643)

Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: khluu <khluu000@gmail.com>
Co-authored-by: jiahao <jxia77@terpmail.umd.edu>
Co-authored-by: Kimi Code <noreply@moonshot.cn>
This commit is contained in:
Jiangyun Zhu
2026-09-08 08:26:07 -07:00
committed by GitHub
co-authored by khluu jiahao Kimi Code
parent 6b15bea080
commit 8ebc5b0a18
5 changed files with 67 additions and 18 deletions
@@ -23,7 +23,7 @@ FP4_DTYPE = torch.uint8
FP8_DTYPE = current_platform.fp8_dtype()
DTYPES = [torch.float16, torch.bfloat16]
SHAPES = [(128, 256), (128, 128), (256, 256), (256, 128)]
SHAPES = [(128, 256), (128, 128), (256, 256), (256, 128), (1, 256), (64, 256)]
BLOCK_SIZE = 16