BSND to TND and FA_UPDATE replacement

Signed-off-by: pichangping <[email protected]>

Author: pichangping

vllm_ascend/attention/attention_v1.py

@@ -917,14 +917,14 @@ def _forward_decode_pcp_dcp(self, query: torch.Tensor,
                                    device=query.device)
 
             graph_params.attn_params[num_tokens].append(
-                (weak_ref_tensors(query), weak_ref_tensors(k_nope), weak_ref_tensors(value), 
-                 self.num_heads, self.num_kv_heads,
+                (weak_ref_tensors(query), weak_ref_tensors(k_nope),
+                 weak_ref_tensors(value), self.num_heads, self.num_kv_heads,
                  self.scale, attn_metadata.block_tables,
                  self.key_cache.shape[1], attn_metadata.decode_meta.
-                 num_computed_tokens_of_pcp_dcp[:, self.pcp_rank, self.dcp_rank],
-                 weak_ref_tensors(workspace), weak_ref_tensors(attn_out), 
-                 weak_ref_tensors(attn_lse), self.pcp_rank, self.dcp_rank,
-                 self.dcp_size))
+                 num_computed_tokens_of_pcp_dcp[:, self.pcp_rank,
+                                                self.dcp_rank],
+                 weak_ref_tensors(attn_out), weak_ref_tensors(attn_lse),
+                 self.pcp_rank, self.dcp_rank, self.dcp_size))
             torch.npu.graph_task_group_begin(stream)
             torch_npu.npu_fused_infer_attention_score.out(
                 query,