解决llm流式返回思考链获取

deepsearcher/agent/deep_search.py

@@ -52,14 +52,14 @@ Query Questions: {query}
 Retrieved Chunks:
 {retrieved_chunks}
 
-Respond with a list of "YES" or "NO" values, one for each chunk, in the same order as the chunks are listed. For example a list of chunks of three: ["YES", "NO", "YES"]"""
+Respond with a list of "YES" or "NO" values, one for each chunk, in the same order as the chunks are listed. For example a list of chunks of three: ["YES", "NO", "YES"]
+"""
 
 
 REFLECT_PROMPT = """
 Determine whether additional search queries are needed based on the original query, previous sub queries, and all retrieved document chunks.
-If further research is required, provide a Python list of up to 3 search queries. If no further research is required, return an empty list.
+If further research is needed (which is prefered, but according to your judge), provide a Python list of more queries.
-
+If no further research is needed, return an empty list.
-If the original query is to write a report, then you prefer to generate some further queries, instead return an empty list.
 
 Original Query: {question}
 

deepsearcher/embedding/openai_embedding.py

@@ -1,5 +1,3 @@
-import os
-
 from openai import OpenAI
 from openai._types import NOT_GIVEN
 
@@ -37,8 +35,6 @@ class OpenAIEmbedding(BaseEmbedding):
 
         if "base_url" in kwargs:
             base_url = kwargs.pop("base_url")
-        else:
-            base_url = os.getenv("OPENAI_BASE_URL")
 
         if "model_name" in kwargs:
             model = kwargs.pop("model_name")

deepsearcher/llm/openai_llm.py

@@ -1,5 +1,3 @@
-import os
-
 from deepsearcher.llm.base import BaseLLM
 
 
@@ -30,12 +28,8 @@ class OpenAILLM(BaseLLM):
         self.model = model
         if "api_key" in kwargs:
             api_key = kwargs.pop("api_key")
-        else:
-            api_key = os.getenv("OPENAI_API_KEY")
         if "base_url" in kwargs:
             base_url = kwargs.pop("base_url")
-        else:
-            base_url = os.getenv("OPENAI_BASE_URL")
         self.client = OpenAI(api_key=api_key, base_url=base_url, **kwargs)
 
     def chat(self, messages: list[dict], stream_callback = None) -> str:
@@ -50,18 +44,27 @@ class OpenAILLM(BaseLLM):
         Returns:
             response (str)
         """
-        completion = self.client.chat.completions.create(
+        with self.client.chat.completions.create(
             model=self.model,
             messages=messages,
-            stream=True
+            stream=True,
-        )
+        ) as stream:
-        response = ""
+            content = ""
-        for chunk in completion:
+            reasoning_content = ""
-            stream_response = chunk.choices[0].delta.content
+            for chunk in stream:
-            if stream_response:
+                if not chunk.choices:
-                print(stream_response, end="", flush=True)
+                    continue
-                response += stream_response
+                else:
-                if stream_callback:
+                    delta = chunk.choices[0].delta
-                    stream_callback(stream_response)
+                if hasattr(delta, 'reasoning_content') and delta.reasoning_content is not None:
+                    print(delta.reasoning_content, end='', flush=True)
+                    reasoning_content += delta.reasoning_content
+                    if stream_callback:
+                        stream_callback(delta.reasoning_content)
+                if hasattr(delta, 'content') and delta.content is not None:
+                    print(delta.content, end="", flush=True)
+                    content += delta.content
+                    if stream_callback:
+                        stream_callback(delta.content)
         print("\n")
-        return response
+        return content