[Perf] [AMD] qwen3: Fuse the 1x128 FP8 input quant into RMSNorm and SiLU-and-mul for block-FP8 linears on gfx950 - #48043

[Perf] [AMD] qwen3: Fuse the 1x128 FP8 input quant into RMSNorm and SiLU-and-mul for block-FP8 linears on gfx950#48043

zengleixin-amd wants to merge 1 commit into

sgl-project:mainsgl-project/sglang:mainfrom

zengleixin-amd:feature/leixin/qwen3-blockfp8-quant-fusionzengleixin-amd/sglang:feature/leixin/qwen3-blockfp8-quant-fusionCopy head branch name to clipboard