[V0 Deprecation] Deprecate BlockSparse Attention & Phi3-Small (#21217)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

vllm/v1/attention/backends/triton_attn.py

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Attention layer with PagedAttention and Triton prefix prefill."""
 from dataclasses import dataclass
-from typing import Any, ClassVar, Optional
+from typing import ClassVar, Optional
 
 import torch
 
@@ -205,15 +205,11 @@ class TritonAttentionImpl(AttentionImpl):
         alibi_slopes: Optional[list[float]],
         sliding_window: Optional[int],
         kv_cache_dtype: str,
-        blocksparse_params: Optional[dict[str, Any]] = None,
         logits_soft_cap: Optional[float] = None,
         attn_type: AttentionType = AttentionType.DECODER,
         kv_sharing_target_layer_name: Optional[int] = None,
         use_irope: bool = False,
     ) -> None:
-        if blocksparse_params is not None:
-            raise ValueError(
-                "TritonAttention does not support block-sparse attention.")
         self.num_heads = num_heads
         self.head_size = head_size
         self.scale = float(scale)