vectorvein 0.1.81__tar.gz → 0.1.83__tar.gz

{vectorvein-0.1.81 → vectorvein-0.1.83}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vectorvein
-Version: 0.1.81
+Version: 0.1.83
 Summary: VectorVein python SDK
 Author-Email: Anderson <andersonby@163.com>
 License: MIT

{vectorvein-0.1.81 → vectorvein-0.1.83}/pyproject.toml

@@ -17,7 +17,7 @@ description = "VectorVein python SDK"
 name = "vectorvein"
 readme = "README.md"
 requires-python = ">=3.10"
-version = "0.1.81"
+version = "0.1.83"
 
 [project.license]
 text = "MIT"

{vectorvein-0.1.81 → vectorvein-0.1.83}/src/vectorvein/chat_clients/openai_compatible_client.py

@@ -3,6 +3,7 @@
 import json
 from functools import cached_property
 from typing import overload, Generator, AsyncGenerator, Any, Literal, Iterable
+import re
 
 import httpx
 from openai import OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI
@@ -229,6 +230,11 @@ class OpenAICompatibleChatClient(BaseChatClient):
                 full_content = ""
                 result = {}
                 usage = None
+                buffer = ""
+                in_reasoning = False
+                current_reasoning = []
+                current_content = []
+
                 for chunk in stream_response:
                     if chunk.usage and chunk.usage.total_tokens:
                         usage = Usage(
@@ -239,14 +245,11 @@ class OpenAICompatibleChatClient(BaseChatClient):
                             completion_tokens_details=chunk.usage.completion_tokens_details,
                         )
 
-                    if len(chunk.choices) == 0:
-                        if usage:
-                            yield ChatCompletionDeltaMessage(usage=usage)
-                        continue
-                    if not chunk.choices[0].delta:
+                    if len(chunk.choices) == 0 or not chunk.choices[0].delta:
                         if usage:
                             yield ChatCompletionDeltaMessage(usage=usage)
                         continue
+
                     if self.model_setting.function_call_available:
                         if chunk.choices[0].delta.tool_calls:
                             for index, tool_call in enumerate(chunk.choices[0].delta.tool_calls):
@@ -254,16 +257,61 @@ class OpenAICompatibleChatClient(BaseChatClient):
                         yield ChatCompletionDeltaMessage(**chunk.choices[0].delta.model_dump(), usage=usage)
                     else:
                         message = chunk.choices[0].delta.model_dump()
-                        full_content += message["content"] if message["content"] else ""
+                        delta_content = message.get("content", "")
+                        buffer += delta_content or ""
+
+                        while True:
+                            if not in_reasoning:
+                                start_pos = buffer.find("<think>")
+                                if start_pos != -1:
+                                    current_content.append(buffer[:start_pos])
+                                    buffer = buffer[start_pos + 7 :]
+                                    in_reasoning = True
+                                else:
+                                    current_content.append(buffer)
+                                    buffer = ""
+                                    break
+                            else:
+                                end_pos = buffer.find("</think>")
+                                if end_pos != -1:
+                                    current_reasoning.append(buffer[:end_pos])
+                                    buffer = buffer[end_pos + 8 :]
+                                    in_reasoning = False
+                                else:
+                                    current_reasoning.append(buffer)
+                                    buffer = ""
+                                    break
+
+                        message["content"] = "".join(current_content).strip()
+                        if current_reasoning:
+                            message["reasoning_content"] = "".join(current_reasoning).strip()
+                        current_content.clear()
+                        current_reasoning.clear()
+
                         if tools:
+                            full_content += message["content"]
                             tool_call_data = ToolCallContentProcessor(full_content).tool_calls
                             if tool_call_data:
                                 message["tool_calls"] = tool_call_data["tool_calls"]
+
                         if full_content in ("<", "<|", "<|▶", "<|▶|") or full_content.startswith("<|▶|>"):
                             message["content"] = ""
                             result = message
                             continue
+
                         yield ChatCompletionDeltaMessage(**message, usage=usage)
+
+                if buffer:
+                    if in_reasoning:
+                        current_reasoning.append(buffer)
+                    else:
+                        current_content.append(buffer)
+                    final_message = {
+                        "content": "".join(current_content).strip(),
+                        "reasoning_content": "".join(current_reasoning).strip() if current_reasoning else None,
+                    }
+                    yield ChatCompletionDeltaMessage(**final_message, usage=usage)
+
                 if result:
                     yield ChatCompletionDeltaMessage(**result, usage=usage)
 
@@ -286,6 +334,13 @@ class OpenAICompatibleChatClient(BaseChatClient):
                 "reasoning_content": getattr(response.choices[0].message, "reasoning_content", None),
                 "usage": response.usage.model_dump() if response.usage else None,
             }
+
+            if not result["reasoning_content"] and result["content"]:
+                think_match = re.search(r"<think>(.*?)</think>", result["content"], re.DOTALL)
+                if think_match:
+                    result["reasoning_content"] = think_match.group(1).strip()
+                    result["content"] = result["content"].replace(think_match.group(0), "", 1).strip()
+
             if tools:
                 if self.model_setting.function_call_available and response.choices[0].message.tool_calls:
                     result["tool_calls"] = [
@@ -501,6 +556,11 @@ class AsyncOpenAICompatibleChatClient(BaseAsyncChatClient):
                 full_content = ""
                 result = {}
                 usage = None
+                buffer = ""
+                in_reasoning = False
+                current_reasoning = []
+                current_content = []
+
                 async for chunk in stream_response:
                     if chunk.usage and chunk.usage.total_tokens:
                         usage = Usage(
@@ -511,11 +571,7 @@ class AsyncOpenAICompatibleChatClient(BaseAsyncChatClient):
                             prompt_tokens_details=chunk.usage.prompt_tokens_details,
                         )
 
-                    if len(chunk.choices) == 0:
-                        if usage:
-                            yield ChatCompletionDeltaMessage(usage=usage)
-                        continue
-                    if not chunk.choices[0].delta:
+                    if len(chunk.choices) == 0 or not chunk.choices[0].delta:
                         if usage:
                             yield ChatCompletionDeltaMessage(usage=usage)
                         continue
@@ -527,16 +583,61 @@ class AsyncOpenAICompatibleChatClient(BaseAsyncChatClient):
                         yield ChatCompletionDeltaMessage(**chunk.choices[0].delta.model_dump(), usage=usage)
                     else:
                         message = chunk.choices[0].delta.model_dump()
-                        full_content += message["content"] if message["content"] else ""
+                        delta_content = message.get("content", "")
+                        buffer += delta_content or ""
+
+                        while True:
+                            if not in_reasoning:
+                                start_pos = buffer.find("<think>")
+                                if start_pos != -1:
+                                    current_content.append(buffer[:start_pos])
+                                    buffer = buffer[start_pos + 7 :]
+                                    in_reasoning = True
+                                else:
+                                    current_content.append(buffer)
+                                    buffer = ""
+                                    break
+                            else:
+                                end_pos = buffer.find("</think>")
+                                if end_pos != -1:
+                                    current_reasoning.append(buffer[:end_pos])
+                                    buffer = buffer[end_pos + 8 :]
+                                    in_reasoning = False
+                                else:
+                                    current_reasoning.append(buffer)
+                                    buffer = ""
+                                    break
+
+                        message["content"] = "".join(current_content).strip()
+                        if current_reasoning:
+                            message["reasoning_content"] = "".join(current_reasoning).strip()
+                        current_content.clear()
+                        current_reasoning.clear()
+
                         if tools:
+                            full_content += message["content"]
                             tool_call_data = ToolCallContentProcessor(full_content).tool_calls
                             if tool_call_data:
                                 message["tool_calls"] = tool_call_data["tool_calls"]
+
                         if full_content in ("<", "<|", "<|▶", "<|▶|") or full_content.startswith("<|▶|>"):
                             message["content"] = ""
                             result = message
                             continue
+
                         yield ChatCompletionDeltaMessage(**message, usage=usage)
+
+                if buffer:
+                    if in_reasoning:
+                        current_reasoning.append(buffer)
+                    else:
+                        current_content.append(buffer)
+                    final_message = {
+                        "content": "".join(current_content).strip(),
+                        "reasoning_content": "".join(current_reasoning).strip() if current_reasoning else None,
+                    }
+                    yield ChatCompletionDeltaMessage(**final_message, usage=usage)
+
                 if result:
                     yield ChatCompletionDeltaMessage(**result, usage=usage)
 
@@ -558,6 +659,13 @@ class AsyncOpenAICompatibleChatClient(BaseAsyncChatClient):
                 "reasoning_content": getattr(response.choices[0].message, "reasoning_content", None),
                 "usage": response.usage.model_dump() if response.usage else None,
             }
+
+            if not result["reasoning_content"] and result["content"]:
+                think_match = re.search(r"<think>(.*?)</think>", result["content"], re.DOTALL)
+                if think_match:
+                    result["reasoning_content"] = think_match.group(1).strip()
+                    result["content"] = result["content"].replace(think_match.group(0), "", 1).strip()
+
             if tools:
                 if self.model_setting.function_call_available and response.choices[0].message.tool_calls:
                     result["tool_calls"] = [