[Perf] Add do_not_specialize in fused FP8 RoPE kernel (#42849)

Signed-off-by: Xin Yang <xyangx@amazon.com>
This commit is contained in:
Xin Yang
2026-05-18 01:32:46 -07:00
committed by GitHub
parent 9537542537
commit 998714b21b
@@ -14,7 +14,7 @@ from vllm.triton_utils import tl, triton
from vllm.utils.torch_utils import direct_register_custom_op
@triton.jit
@triton.jit(do_not_specialize=["num_tokens"])
def _fused_inv_rope_fp8_quant_per_head(
o_ptr,
positions_ptr,