Fuse linear projections, SiLU activation, and replace conv1d to reduce kernel launches#18392

mergennachin merged 1 commit intomainpytorch/executorch:mainfrom

mnachin/qwen3_5_moe_fused_experts_v2pytorch/executorch:mnachin/qwen3_5_moe_fused_experts_v2Copy head branch name to clipboard