Hi!
I want to use the Matmul operator for custom matrix multiplication, however, it is only supported with Multi Head Attention (according to documentation).
Are there plans to change it soon or do you have some tips on how to solve it differently?
Hi @mateusz.piechocki,
Currently we only support the MHAT as you pointed out. We plan to implement support for other Matmul version as well, but there is no specific timeline for it.
Regards,