[1/n]: add cutlass W4A8 moe kernel for hopper architecture (#7772)

Signed-off-by: yangsijia.614 <yangsijia.614@bytedance.com> Co-authored-by: yicwang <yichen.wang@bytedance.com>
2025-07-05 11:50:12 +08:00
parent cb432f1770
commit da3890e82a
16 changed files with 3602 additions and 0 deletions
--- a/sgl-kernel/python/sgl_kernel/init.py
+++ b/sgl-kernel/python/sgl_kernel/init.py
@@ -19,6 +19,7 @@ from sgl_kernel.attention import (
    merge_state,
    merge_state_v2,
 )
+from sgl_kernel.cutlass_moe import cutlass_w4a8_moe_mm, get_cutlass_w4a8_moe_mm_data
 from sgl_kernel.elementwise import (
    apply_rope_with_cos_sin_cache_inplace,
    fused_add_rmsnorm,