GitLab

[Kernel] optimize moe_align_block_size for cuda graph and large num_experts (e.g. DeepSeek-V3) (#12222)
Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>
Co-authored-by: Michael Goin <mgoin@redhat.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>

Signed-off-by: mgoin <michael@neuralmagic.com>
Co-authored-by: mgoin <michael@neuralmagic.com>
Co-authored-by: robertgshaw2-neuralmagic <rshaw@neuralmagic.com>

Signed-off-by: charlifu <charlifu@amd.com>

Co-authored-by: Dipika <dipikasikka1@gmail.com>
Co-authored-by: Dipika Sikka <ds3822@columbia.edu>

Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>

Co-authored-by: mgoin <michael@neuralmagic.com>

Co-authored-by: lwilkinson@neuralmagic.com

Co-authored-by: Dipika <dipikasikka1@gmail.com>

Co-authored-by: ElizaWszola <eliza@neuralmagic.com>

Co-authored-by: ElizaWszola <eliza@neuralmagic.com>

This PR enables the fused topk_softmax kernel used in moe layer for HIP