feat: add think display for volcengine and generic openapi (#13234)

Signed-off-by: xhe <xw897002528@gmail.com>

api/core/model_runtime/model_providers/openai_api_compatible/llm/llm.py

@@ -473,6 +473,8 @@ class OAIAPICompatLargeLanguageModel(_CommonOaiApiCompat, LargeLanguageModel):
 
         finish_reason = None  # The default value of finish_reason is None
         message_id, usage = None, None
+        is_reasoning_started = False
+        is_reasoning_started_tag = False
         for chunk in response.iter_lines(decode_unicode=True, delimiter=delimiter):
             chunk = chunk.strip()
             if chunk:
@@ -514,6 +516,33 @@ class OAIAPICompatLargeLanguageModel(_CommonOaiApiCompat, LargeLanguageModel):
                     delta = choice["delta"]
                     delta_content = delta.get("content")
 
+                    if not is_reasoning_started_tag and "<think>" in delta_content:
+                        is_reasoning_started_tag = True
+                        delta_content = "> 💭 " + delta_content.replace("<think>", "")
+                    elif is_reasoning_started_tag and "</think>" in delta_content:
+                        delta_content = delta_content.replace("</think>", "") + "\n\n"
+                        is_reasoning_started_tag = False
+                    elif is_reasoning_started_tag:
+                        if "\n\n" in delta_content:
+                            delta_content = delta_content.replace("\n\n", "\n> ")
+                        elif "\n" in delta_content:
+                            delta_content = delta_content.replace("\n", "\n> ")
+
+                    reasoning_content = delta.get("reasoning_content")
+                    if reasoning_content:
+                        if not is_reasoning_started:
+                            delta_content = "> 💭 " + reasoning_content
+                            is_reasoning_started = True
+                        elif "\n\n" in delta_content:
+                            delta_content = reasoning_content.replace("\n\n", "\n> ")
+                        elif "\n" in delta_content:
+                            delta_content = reasoning_content.replace("\n", "\n> ")
+                    elif is_reasoning_started:
+                        # If we were in reasoning mode but now getting regular content,
+                        # add \n\n to close the reasoning block
+                        delta_content = "\n\n" + delta_content
+                        is_reasoning_started = False
+
                     assistant_message_tool_calls = None
 
                     if "tool_calls" in delta and credentials.get("function_calling_type", "no_call") == "tool_call":

api/core/model_runtime/model_providers/volcengine_maas/llm/llm.py

@@ -247,15 +247,36 @@ class VolcengineMaaSLargeLanguageModel(LargeLanguageModel):
             req_params["tools"] = tools
 
         def _handle_stream_chat_response(chunks: Generator[ChatCompletionChunk]) -> Generator:
+            is_reasoning_started = False
             for chunk in chunks:
+                content = ""
+                if chunk.choices:
+                    if hasattr(chunk.choices[0].delta, "reasoning_content"):
+                        delta_content = ""
+                        if not is_reasoning_started:
+                            is_reasoning_started = True
+                            delta_content = "> 💭 " + chunk.choices[0].delta.reasoning_content
+                        else:
+                            delta_content = chunk.choices[0].delta.reasoning_content
+
+                        if "\n\n" in delta_content:
+                            delta_content = delta_content.replace("\n\n", "\n> ")
+                        elif "\n" in delta_content:
+                            delta_content = delta_content.replace("\n", "\n> ")
+
+                        content = delta_content
+                    elif is_reasoning_started:
+                        content = "\n\n" + chunk.choices[0].delta.content
+                        is_reasoning_started = False
+                    else:
+                        content = chunk.choices[0].delta.content
+
                 yield LLMResultChunk(
                     model=model,
                     prompt_messages=prompt_messages,
                     delta=LLMResultChunkDelta(
                         index=0,
-                        message=AssistantPromptMessage(
-                            content=chunk.choices[0].delta.content if chunk.choices else "", tool_calls=[]
-                        ),
+                        message=AssistantPromptMessage(content=content, tool_calls=[]),
                         usage=self._calc_response_usage(
                             model=model,
                             credentials=credentials,