Simplify flashinfer dispatch (#1552)

python/sglang/srt/layers/radix_attention.py

@@ -32,9 +32,10 @@ class RadixAttention(nn.Module):
         scaling: float,
         num_kv_heads: int,
         layer_id: int,
-        sliding_window_size: int = -1,
         logit_cap: float = 0.0,
         v_head_dim: int = -1,
+        sliding_window_size: int = -1,
+        is_cross_attention: bool = False,
     ):
         super().__init__()
         self.tp_q_head_num = num_heads
@@ -47,6 +48,7 @@ class RadixAttention(nn.Module):
         self.layer_id = layer_id
         self.logit_cap = logit_cap
         self.sliding_window_size = sliding_window_size or -1
+        self.is_cross_attention = is_cross_attention
 
     def forward(self, q, k, v, forward_batch: ForwardBatch):
         if k is not None: