通过llm组件按钮控制是否流式输出，技能控制流式而非配置控制流式

src/backend/bisheng/chat/utils.py

@@ -5,7 +5,6 @@
 from bisheng.database.base import session_getter
 from bisheng.database.models.model_deploy import ModelDeploy
 from bisheng.database.models.recall_chunk import RecallChunk
-from bisheng.interface.utils import try_setting_streaming_options
 from bisheng.processing.base import get_result_and_steps
 from bisheng.utils.logger import logger
 from bisheng_langchain.chat_models import HostQwenChat
@@ -21,7 +20,6 @@ async def process_graph(langchain_object,
                         websocket: WebSocket,
                         flow_id: str = None,
                         chat_id: str = None):
-    langchain_object = try_setting_streaming_options(langchain_object, websocket)
     logger.debug('Loaded langchain object')
 
     if langchain_object is None:

src/backend/bisheng/interface/initialize/loading.py

@@ -2,7 +2,6 @@
 from typing import Any, Callable, Dict, Sequence, Type
 
 from bisheng.cache.utils import file_download
-from bisheng.chat.config import ChatConfig
 from bisheng.interface.agents.base import agent_creator
 from bisheng.interface.chains.base import chain_creator
 from bisheng.interface.custom_lists import CUSTOM_NODES
@@ -24,7 +23,6 @@
 from langchain.agents.agent import AgentExecutor
 from langchain.agents.agent_toolkits.base import BaseToolkit
 from langchain.agents.tools import BaseTool
-from langchain.base_language import BaseLanguageModel
 from langchain.chains.base import Chain
 from langchain.document_loaders.base import BaseLoader
 from langchain.vectorstores.base import VectorStore
@@ -214,18 +212,10 @@ def instantiate_llm(node_type, class_object, params: Dict):
             params['max_tokens'] = int(params['max_tokens'])
         elif not isinstance(params.get('max_tokens'), int):
             params.pop('max_tokens', None)
-    # 支持stream
-    llm = class_object(**params)
-    llm_config = settings.get_from_db('llm_request')
-    if isinstance(llm, BaseLanguageModel):
-        if hasattr(llm, 'streaming') and isinstance(llm.streaming, bool):
-            llm.streaming = llm_config.get(
-                'stream') if 'stream' in llm_config else ChatConfig.streaming
-        elif hasattr(llm, 'stream') and isinstance(llm.stream, bool):
-            llm.stream = llm_config.get(
-                'stream') if 'stream' in llm_config else ChatConfig.streaming
 
     # 支持request_timeout & max_retries
+    llm = class_object(**params)
+    llm_config = settings.get_from_db('llm_request')
     if hasattr(llm, 'request_timeout') and 'request_timeout' in llm_config:
         if isinstance(llm_config.get('request_timeout'), str):
             llm.request_timeout = eval(llm_config.get('request_timeout'))

src/backend/bisheng/interface/utils.py

@@ -5,10 +5,7 @@
 from io import BytesIO
 
 import yaml
-from bisheng.chat.config import ChatConfig
-from bisheng.settings import settings
 from bisheng.utils.logger import logger
-from langchain.base_language import BaseLanguageModel
 from PIL.Image import Image
 
 
@@ -35,28 +32,6 @@ def pil_to_base64(image: Image) -> str:
     return img_str.decode('utf-8')
 
 
-def try_setting_streaming_options(langchain_object, websocket):
-    # If the LLM type is OpenAI or ChatOpenAI,
-    # set streaming to True
-    # First we need to find the LLM
-    llm = None
-    if hasattr(langchain_object, 'llm'):
-        llm = langchain_object.llm
-    elif hasattr(langchain_object, 'llm_chain') and hasattr(langchain_object.llm_chain, 'llm'):
-        llm = langchain_object.llm_chain.llm
-
-    if isinstance(llm, BaseLanguageModel):
-        if hasattr(llm, 'streaming') and isinstance(llm.streaming, bool):
-            llm.streaming = settings.get_from_db('llm_request').get(
-                'stream') if 'stream' in settings.get_from_db(
-                    'llm_request') else ChatConfig.streaming
-        elif hasattr(llm, 'stream') and isinstance(llm.stream, bool):
-            llm.stream = settings.get_from_db('llm_request').get(
-                'stream') if 'stream' in settings.get_from_db(
-                    'llm_request') else ChatConfig.streaming
-    return langchain_object
-
-
 def extract_input_variables_from_prompt(prompt: str) -> list[str]:
     """Extract input variables from prompt."""
     return re.findall(r'{(.*?)}', prompt)

src/backend/bisheng/services/chat/utils.py

@@ -1,5 +1,4 @@
 from bisheng.api.v1.schemas import ChatMessage
-from bisheng.interface.utils import try_setting_streaming_options
 from bisheng.processing.base import get_result_and_steps
 from loguru import logger
 
@@ -10,7 +9,6 @@ async def process_graph(
     client_id: str,
     session_id: str,
 ):
-    langchain_object = try_setting_streaming_options(langchain_object)
     logger.debug('Loaded langchain object')
 
     if langchain_object is None:

src/backend/bisheng/template/frontend_node/llms.py

@@ -114,6 +114,7 @@ def format_contribute_field(field: TemplateField):
             'top_p',
             'top_k',
             'max_tokens',
+            'streaming'
         ]
         if field.name in show_fields:
             field.show = True