move enable_helix

Author: brb-nv

tensorrt_llm/_torch/attention_backend/trtllm.py

@@ -702,6 +702,8 @@ def host_kv_cache_pool_mapping(self) -> Optional[torch.Tensor]:
 
     def __post_init__(self) -> None:
         super().__post_init__()
+        self.enable_helix = self.mapping.has_cp_helix(
+        ) if self.mapping is not None else False
         self._post_init_with_buffers(self.cuda_graph_buffers)
 
     def _post_init_with_buffers(self, buffers) -> None:

tensorrt_llm/_torch/pyexecutor/model_engine.py

@@ -1037,8 +1037,6 @@ def _set_up_attn_metadata(self, kv_cache_manager: KVCacheManager):
             cache_indirection=cache_indirection,
             sparse_attention_config=self.sparse_attention_config,
             num_heads_per_kv=num_heads_per_kv,
-            enable_helix=self.mapping.has_cp_helix()
-            if self.mapping is not None else False,
         )
 
         return self.attn_metadata