pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2026-02-02 00:25:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refa: asyncio.to_thread to ThreadPoolExecutor to break thread limitat… (#12716)

Browse Source 
### Type of change

- [x] Refactoring
This commit is contained in:
Kevin Hu
2026-01-20 13:29:37 +08:00
committed by GitHub
parent 120648ac81
commit 927db0b373
30 changed files with 246 additions and 157 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
rag/llm/chat_model.py
View File

@ -34,8 +34,9 @@ from common.token_utils import num_tokens_from_string, total_token_count_from_re
from rag.llm import FACTORY_DEFAULT_BASE_URL, LITELLM_PROVIDER_PREFIX, SupportedLiteLLMProvider
from rag.nlp import is_chinese, is_english
# Error message constants
from common.misc_utils import thread_pool_exec
class LLMErrorCode(StrEnum):
ERROR_RATE_LIMIT = "RATE_LIMIT_EXCEEDED"
ERROR_AUTHENTICATION = "AUTH_ERROR"
@ -309,7 +310,7 @@ class Base(ABC):
name = tool_call.function.name
try:
args = json_repair.loads(tool_call.function.arguments)
tool_response = await asyncio.to_thread(self.toolcall_session.tool_call, name, args)
tool_response = await thread_pool_exec(self.toolcall_session.tool_call, name, args)
history = self._append_history(history, tool_call, tool_response)
ans += self._verbose_tool_use(name, args, tool_response)
except Exception as e:
@ -402,7 +403,7 @@ class Base(ABC):
try:
args = json_repair.loads(tool_call.function.arguments)
yield self._verbose_tool_use(name, args, "Begin to call...")
tool_response = await asyncio.to_thread(self.toolcall_session.tool_call, name, args)
tool_response = await thread_pool_exec(self.toolcall_session.tool_call, name, args)
history = self._append_history(history, tool_call, tool_response)
yield self._verbose_tool_use(name, args, tool_response)
except Exception as e:
@ -1462,7 +1463,7 @@ class LiteLLMBase(ABC):
name = tool_call.function.name
try:
args = json_repair.loads(tool_call.function.arguments)
tool_response = await asyncio.to_thread(self.toolcall_session.tool_call, name, args)
tool_response = await thread_pool_exec(self.toolcall_session.tool_call, name, args)
history = self._append_history(history, tool_call, tool_response)
ans += self._verbose_tool_use(name, args, tool_response)
except Exception as e:
@ -1562,7 +1563,7 @@ class LiteLLMBase(ABC):
try:
args = json_repair.loads(tool_call.function.arguments)
yield self._verbose_tool_use(name, args, "Begin to call...")
tool_response = await asyncio.to_thread(self.toolcall_session.tool_call, name, args)
tool_response = await thread_pool_exec(self.toolcall_session.tool_call, name, args)
history = self._append_history(history, tool_call, tool_response)
yield self._verbose_tool_use(name, args, tool_response)
except Exception as e: