Cunxiao Ni
Cunxiao Ni
The compilation error is caused by the absence of a direct bf16-to-fp8 cast function in cutlass's float8.h. To resolve this issue, an intermediate conversion is required: bf16 should first be...
ok, I will take a look.
It seems there are still some issues when the buffer is 2D...
@LeiWang1999 ok, I will.
@oraluben It seems `pass_config` can also accept a parameter of a specific type. In that case, I think we should stop using `True` or `False` for the check and just...
@oraluben fixed==
@jiangwenj02 Hi, can you provide more information about your examples and environment? I can't reproduce this error on A100.