[Benchmarks] Fix imports in FP8 tuning script (#26407)

Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com>

benchmarks/kernels/benchmark_w8a8_block_fp8.py

@@ -14,7 +14,7 @@ import torch
 from tqdm import tqdm
 
 from vllm.model_executor.layers.quantization.utils.fp8_utils import (
-    _w8a8_block_fp8_matmul,
+    _w8a8_triton_block_scaled_mm,
 )
 from vllm.platforms import current_platform
 from vllm.triton_utils import triton
@@ -83,7 +83,7 @@ def w8a8_block_matmul(
         )
 
     if A.dtype == torch.float8_e4m3fn:
-        kernel = _w8a8_block_fp8_matmul
+        kernel = _w8a8_triton_block_scaled_mm
     else:
         raise RuntimeError("Currently, only support tune w8a8 block fp8 kernel.")