Fix runtime bug

Signed-off-by: Jiayu Chang <jiayuc@nvidia.com>

Author: JyChang012

tensorrt_llm/_torch/pyexecutor/model_engine.py

@@ -2667,7 +2667,7 @@ def previous_seq_slots_device():
 
         attn_metadata.prepare()
 
-        peft_cache_manager = resource_manager.get_resource_manager(
+        peft_cache_manager = resource_manager and resource_manager.get_resource_manager(
             ResourceManagerType.PEFT_CACHE_MANAGER)
         lora_params = self._get_lora_params_from_requests(
             scheduled_requests, attn_metadata, peft_cache_manager, maybe_graph)
@@ -3307,8 +3307,6 @@ def forward(self,
                 req_id_to_old_request: Optional[Dict[int, LlmRequest]] = None):
         kv_cache_manager = resource_manager.get_resource_manager(
             self.kv_cache_manager_key)
-        peft_cache_manager = resource_manager.get_resource_manager(
-            ResourceManagerType.PEFT_CACHE_MANAGER)
 
         attn_metadata = self._set_up_attn_metadata(kv_cache_manager)
         if self.enable_spec_decode: