[sgl-kernel][1/2] Fused qk_norm_rope for GLM4.6 (#15141)

2025-12-18 17:07:04 +08:00
parent fea2d5211d
commit 4792d1f452
5 changed files with 77 additions and 43 deletions
--- a/sgl-kernel/include/sgl_kernel_ops.h
+++ b/sgl-kernel/include/sgl_kernel_ops.h
@@ -401,7 +401,8 @@ void fused_qk_norm_rope(
    double factor,
    double low,
    double high,
-    double attention_factor);
+    double attention_factor,
+    int64_t rotary_dim);

 void cutlass_fp4_group_mm(
    torch::Tensor& output,