[Intel HPU] enable level 1 prefix caching and fix some bugs (#4971)

* [Intel HPU] enable prefix caching and dense tp moe ep and fix some bugs

* update code by copilot

* remove dense tp and moe ep code

fastdeploy/engine/engine.py

@@ -40,6 +40,7 @@ from fastdeploy.engine.expert_service import start_data_parallel_service
 from fastdeploy.engine.request import Request
 from fastdeploy.inter_communicator import EngineWorkerQueue, IPCSignal
 from fastdeploy.metrics.metrics import main_process_metrics
+from fastdeploy.platforms import current_platform
 from fastdeploy.utils import EngineError, console_logger, envs, llm_logger
 
 
@@ -136,8 +137,9 @@ class LLMEngine:
 
         # If block numer is specified and model is deployed in mixed mode, start cache manager first
         if not self.do_profile and self.cfg.scheduler_config.splitwise_role != "mixed":
-            device_ids = self.cfg.parallel_config.device_ids.split(",")
-            self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix)
+            if not current_platform.is_intel_hpu():
+                device_ids = self.cfg.parallel_config.device_ids.split(",")
+                self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix)
 
         # Start workers
         self.worker_proc = self._start_worker_service()
@@ -170,8 +172,9 @@ class LLMEngine:
         if self.do_profile:
             self._stop_profile()
         elif self.cfg.scheduler_config.splitwise_role == "mixed" and self.cfg.cache_config.enable_prefix_caching:
-            device_ids = self.cfg.parallel_config.device_ids.split(",")
-            self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix)
+            if not current_platform.is_intel_hpu():
+                device_ids = self.cfg.parallel_config.device_ids.split(",")
+                self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix)
 
         # Launch components: scheduler, cache_manager, expert_service et.al.
         if self.cfg.scheduler_config.splitwise_role != "mixed":
@@ -673,8 +676,9 @@ class LLMEngine:
         self.cfg.cache_config.reset(num_gpu_blocks)
         self.engine.resource_manager.reset_cache_config(self.cfg.cache_config)
         if self.cfg.cache_config.enable_prefix_caching or self.cfg.scheduler_config.splitwise_role != "mixed":
-            device_ids = self.cfg.parallel_config.device_ids.split(",")
-            self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix)
+            if not current_platform.is_intel_hpu():
+                device_ids = self.cfg.parallel_config.device_ids.split(",")
+                self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix)
 
     def check_health(self, time_interval_threashold=30):
         """