Support FA3 as Attention backend by using --attention-backend fa3 (#4680)

Co-authored-by: qsong <qsong@linkedin.com>
Co-authored-by: qingquansong <ustcsqq@gmail.com>

python/sglang/bench_serving.py

@@ -501,6 +501,7 @@ def get_dataset(args, tokenizer):
             question_len=args.gsp_question_len,
             output_len=args.gsp_output_len,
             tokenizer=tokenizer,
+            args=args,
         )
     else:
         raise ValueError(f"Unknown dataset: {args.dataset_name}")
@@ -788,6 +789,7 @@ def sample_generated_shared_prefix_requests(
     question_len: int,
     output_len: int,
     tokenizer: PreTrainedTokenizerBase,
+    args: argparse.Namespace,
 ) -> List[Tuple[str, int, int]]:
     """Generate benchmark requests with shared system prompts using random tokens and caching."""
     cache_path = get_gen_prefix_cache_path(args, tokenizer)