updated

vllm/v1/engine/async_llm_engine.py

@@ -268,45 +268,3 @@ async def run_output_handler(self):
             # List[RequestOutput] rather than pushing to the Queue at the
             # expense of doing another loop through List[RequestOutput] here.
             self.detokenizer.step_streaming(engine_core_outputs)
-
-    async def abort(self):
-        pass
-
-    async def check_health(self):
-        pass
-
-    async def dead_error(self):
-        pass
-
-    async def do_log_stats(self):
-        pass
-
-    async def encode(self):
-        pass
-
-    async def errored(self):
-        pass
-
-    async def get_decoding_config(self):
-        pass
-
-    async def get_model_config(self):
-        pass
-
-    async def get_tokenizer(self):
-        pass
-
-    async def is_running(self):
-        pass
-
-    async def is_stopped(self):
-        pass
-
-    async def is_tracing_enabled(self):
-        pass
-
-    async def start_profile(self):
-        pass
-
-    async def stop_profile(self):
-        pass

vllm/v1/engine/llm_engine.py

@@ -17,13 +17,14 @@
 from vllm.usage.usage_lib import UsageContext
 from vllm.v1.engine.detokenizer import Detokenizer
 from vllm.v1.engine.llm_engine_core import LLMEngineCore
+from vllm.v1.engine.protocol import LLMEngineProtocol
 from vllm.v1.engine.processor import Processor
 from vllm.v1.executor.gpu_executor import GPUExecutor
 
 logger = init_logger(__name__)
 
 
-class LLMEngine:
+class LLMEngine(LLMEngineProtocol):
 
     def __init__(
         self,
@@ -141,9 +142,3 @@ def step(self) -> List[RequestOutput]:
         request_outputs = self.detokenizer.step(engine_core_outputs)
 
         return request_outputs
-
-    def get_num_unfinished_requests(self) -> int:
-        return self.detokenizer.get_num_unfinished_requests()
-
-    def has_unfinished_requests(self) -> bool:
-        return self.detokenizer.has_unfinished_requests()