updated

vllm/v1/engine/async_llm.py

@@ -1,6 +1,6 @@
 import asyncio
 import os
 from typing import AsyncGenerator, Dict, List, Mapping, Optional, Type, Union
 
 from vllm.config import ModelConfig, VllmConfig
 from vllm.engine.arg_utils import AsyncEngineArgs
@@ -20,7 +20,7 @@
 from vllm.v1.engine.core_client import EngineCoreClient
 from vllm.v1.engine.output_processor import OutputProcessor
 from vllm.v1.engine.processor import Processor
 from vllm.v1.engine.request_state import RequestState
 from vllm.v1.executor.abstract import Executor
 from vllm.v1.metrics.loggers import LoggingStatLogger, StatLoggerBase
 from vllm.v1.metrics.stats import IterationStats, SchedulerStats
@@ -235,19 +235,17 @@
                 engine_core_outputs = await self.engine_core.get_output_async()
 
                 # 2) Process EngineCoreOutputs.
-                processed_outputs = self.output_processor.process_outputs(
-                    engine_core_outputs, self.request_states)
+                outputs = self.output_processor.process_outputs(engine_core_outputs)
 
                 # 3) Abort any reqs that finished due to stop strings.
-                await self.engine_core.abort_requests_async(
-                    processed_outputs.reqs_to_abort)
+                await self.engine_core.abort_requests_async(outputs.reqs_to_abort)
 
                 # 4) Logging.
                 # TODO(rob): make into a coroutine and launch it in
                 # background thread once we add Prometheus.
                 self._log_stats(
                     scheduler_stats=engine_core_outputs.scheduler_stats,
-                    iteration_stats=processed_outputs.iteration_stats,
+                    iteration_stats=outputs.iteration_stats,
                 )
 
         except Exception as e:
@@ -261,7 +259,6 @@
         await self.engine_core.abort_requests_async(request_ids)
         self.output_processor.abort_requests(request_ids)
 
-
     def _log_stats(
         self,
         scheduler_stats: SchedulerStats,

vllm/v1/engine/output_processor.py

@@ -112,7 +112,6 @@ def make_request_output(
     def process_outputs(
         self,
         outputs: EngineCoreOutputs,
-        request_states: Dict[str, RequestState],
     ) -> OutputProcessorOutput:
         """
         Process the EngineCoreOutputs:
@@ -146,7 +145,7 @@ def process_outputs(
         iteration_stats = IterationStats(self.log_stats)
         for engine_core_output in outputs.outputs:
             req_id = engine_core_output.request_id
-            req_state = request_states.get(req_id)
+            req_state = self.request_states.get(req_id)
             if req_state is None:
                 # Ignore output for already-aborted request.
                 continue
@@ -173,7 +172,7 @@ def process_outputs(
 
                 # Free completed requests.
                 if request_output.finished:
-                    request_states.pop(req_id)
+                    self.request_states.pop(req_id)
                     if not engine_core_output.finished:
                         # If req not finished in EngineCore, but Detokenizer
                         # detected stop string, abort needed in EngineCore.