[Engine] [Feature] Refactor async_llm:cross-process with EngineService,based on zmq communication (#4868)

* Refactor async_llm:cross-process with EngineService

* fix: async_llm output process

* fix: return prompt_token_ids and prompt_tokens in first res

* optimize common_engine start func
This commit is contained in:
zhouchong
2025-12-09 10:53:40 +08:00
committed by GitHub
parent 2f208db4e9
commit 5d9b5e4a5b
8 changed files with 2217 additions and 1790 deletions
+1 -1
View File
@@ -58,7 +58,7 @@ class BaseDataProcessor(ABC):
def set_value(req, key, value):
value = getattr(self.generation_config, key, value)
if isinstance(req, dict):
if key not in req:
if key not in req or req[key] is None:
req[key] = value
else:
if req.get(key) is None: