[fix] fix prefix caching not enabled

fastdeploy/cache_manager/prefix_cache_manager.py

@@ -230,7 +230,7 @@ class PrefixCacheManager:
         while np.sum(self.cache_ready_signal.value) != tensor_parallel_size:
             time.sleep(1)
 
-        if cache_config.swap_space is not None and cache_config.swap_space > 0:
+        if cache_config.enable_hierarchical_cache and self.num_cpu_blocks > 0:
             while np.sum(self.swap_space_ready_signal.value) != tensor_parallel_size:
                 time.sleep(1)
 

fastdeploy/engine/args_utils.py

@@ -385,6 +385,7 @@ class EngineArgs:
         """
         Post-initialization processing to set default tokenizer if not provided.
         """
+        from fastdeploy.utils import llm_logger
         if not self.tokenizer:
             self.tokenizer = self.model
         if self.enable_logprob:

fastdeploy/engine/engine.py

@@ -163,15 +163,7 @@ class LLMEngine:
             self._stop_profile()
         elif self.cfg.cache_config.enable_prefix_caching:
             device_ids = self.cfg.device_ids.split(",")
-            self.cache_manager_processes = self.resource_manager.cache_manager.launch_cache_manager(
-                cache_config=self.cfg.cache_config,
-                tensor_parallel_size=self.cfg.tensor_parallel_size,
-                device_ids=device_ids,
-                pod_ip=self.cfg.master_ip,
-                engine_worker_queue_port=self.cfg.engine_worker_queue_port,
-                pid_suffix=self.ipc_signal_suffix,
-                create_cache_tensor=False,
-            )
+            self.cache_manager_processes = self.engine.start_cache_service(device_ids, self.ipc_signal_suffix, False)
 
         # Launch components: scheduler, cache_manager, expert_service et.al.
         self.launch_components()