Set baseline for ragged_attention (#284)

Author: yf225

tritonbench/operators/ragged_attention/operator.py

@@ -74,7 +74,7 @@ def __init__(
         self.alpha = 1.0 / self.attn_dim
         self.requires_grad = not (self.mode == Mode.FWD_NO_GRAD)
 
-    @register_benchmark()
+    @register_benchmark(baseline=True)
     def hstu(self, q, k, v, seq_offsets, num_targets, max_seq_len):
         return lambda: triton_hstu_mha(
             max_seq_len,