Fix: Properly handle tool calls with results returned to API

This commit is contained in:
JianFeeeee
2026-04-11 22:54:02 +08:00
parent 57697e464b
commit f577967489

View File

@ -1,8 +1,9 @@
"""聊天服务"""
import asyncio
import json
from datetime import datetime
from typing import AsyncIterator, TYPE_CHECKING
from typing import AsyncIterator, TYPE_CHECKING, List, Dict, Any
from ..core.imports import GraphMemoryClient
from ..models.message import ToolCall, ToolResult
from .tool_service import ToolService
@ -23,7 +24,7 @@ class ChatService:
self._graph = graph
self._client = client
self._tool_service = tool_service
self._messages: list[dict] = []
self._messages: List[Dict[str, Any]] = []
async def send_message(self, user_input: str) -> AsyncIterator[dict]:
"""发送消息并流式返回事件"""
@ -34,13 +35,12 @@ class ChatService:
}
try:
# 2. 使用流式API调用
# 2. 第一次API调用
accumulated_content = ""
tool_calls_data = []
# 流式处理响应
async for chunk in self._call_api_stream_async(user_input):
# 处理内容增量
if chunk.get("content_delta"):
accumulated_content += chunk["content_delta"]
yield {
@ -48,11 +48,10 @@ class ChatService:
"content": accumulated_content
}
# 处理工具调用
if chunk.get("tool_calls"):
tool_calls_data = chunk["tool_calls"]
# 3. 处理工具调用
# 3. 如果有工具调用执行并继续调用API
tool_calls = None
tool_results = None
@ -60,26 +59,23 @@ class ChatService:
tool_calls = []
tool_results = []
# 执行所有工具
for tool_call_data in tool_calls_data:
# 创建工具调用对象
tool_call = ToolCall(
id=tool_call_data.id,
name=tool_call_data.function.name,
arguments=tool_call_data.function.arguments
id=tool_call_data["id"],
name=tool_call_data["function"]["name"],
arguments=tool_call_data["function"]["arguments"]
)
tool_calls.append(tool_call)
# 发送工具调用事件
yield {
"type": "tool_call",
"tool_call": tool_call
}
# 执行工具
result = await self._tool_service.execute(tool_call)
tool_results.append(result)
# 发送工具结果事件
log_entry = ToolService._create_log_entry(tool_call, result)
yield {
"type": "tool_result",
@ -87,6 +83,47 @@ class ChatService:
"log_entry": log_entry
}
# 构建工具结果消息
tool_messages = []
for tc, tr in zip(tool_calls, tool_results):
tool_messages.append({
"role": "tool",
"tool_call_id": tc.id,
"content": tr.content
})
# 构建assistant消息包含tool_calls
assistant_message = {
"role": "assistant",
"content": accumulated_content,
"tool_calls": [
{
"id": tc.id,
"type": "function",
"function": {
"name": tc.name,
"arguments": tc.arguments
}
} for tc in tool_calls
]
}
# 第二次API调用传入工具结果
final_content = ""
async for chunk in self._call_api_stream_with_tools(
user_input,
assistant_message,
tool_messages
):
if chunk.get("content_delta"):
final_content += chunk["content_delta"]
yield {
"type": "content_delta",
"content": final_content
}
accumulated_content = final_content
# 4. 返回最终回复
yield {
"type": "assistant_message",
@ -96,7 +133,6 @@ class ChatService:
}
except Exception as e:
# 错误处理
yield {
"type": "error",
"error": str(e)
@ -106,7 +142,6 @@ class ChatService:
"""异步流式调用 API"""
loop = asyncio.get_event_loop()
# 在executor中运行同步流式API
def process_stream():
stream = self._client.send_message_stream(message)
tool_calls_accumulated = []
@ -114,14 +149,11 @@ class ChatService:
for chunk in stream:
delta = chunk.choices[0].delta
# 处理内容增量
if delta.content:
yield {"content_delta": delta.content}
# 处理工具调用
if delta.tool_calls:
for tc in delta.tool_calls:
# 累积工具调用数据
if tc.index >= len(tool_calls_accumulated):
tool_calls_accumulated.append({
"id": tc.id,
@ -138,11 +170,41 @@ class ChatService:
if tc.function.arguments:
tool_calls_accumulated[tc.index]["function"]["arguments"] += tc.function.arguments
# 返回完整的工具调用
if tool_calls_accumulated:
yield {"tool_calls": tool_calls_accumulated}
# 使用run_in_executor处理生成器
for result in await loop.run_in_executor(None, lambda: list(process_stream())):
yield result
async def _call_api_stream_with_tools(
self,
user_input: str,
assistant_message: dict,
tool_messages: list
) -> AsyncIterator[dict]:
"""带工具结果的流式调用"""
loop = asyncio.get_event_loop()
def process_stream():
# 构建完整的消息列表
messages = [
{"role": "user", "content": user_input},
assistant_message
]
messages.extend(tool_messages)
# 调用API
response = self._client.client.chat.completions.create(
model=self._client.tools[0]["function"]["name"] if self._client.tools else "deepseek-chat",
messages=messages,
stream=True
)
for chunk in response:
delta = chunk.choices[0].delta
if delta.content:
yield {"content_delta": delta.content}
for result in await loop.run_in_executor(None, lambda: list(process_stream())):
yield result