def fix_json(json_string, round_count = 0):
    round_count += 1
    try:
        json.loads(json_string)
        return json_string
    except json.JSONDecodeError as e:
        print("[Worker Agent Activated]: Round", round_count)
        print("Fix JSON Format Error:\n", e.msg)
        print("Origin String:\n", json_string, "\n")
        fixed_result = worker\
            .input({
                "origin JSON String": json_string,
                "error": e.msg,
                "position": e.pos,
            })\
            .output("Fixed JSON String only without explanation and decoration.")\
            .start()
        print("Fixed Content:\n", fixed_result, "\n")
        return fix_json(fixed_result, round_count)

result = fix_json("{'words': ['apple', 'banana', 'carrot', 'dog', 'elephant'], 'sentence': 'I have an apple, a banana, a carrot, a dog, and an elephant.'}")
print(result)

# 首先我们定义一下可用的工具
tools = {
    "weather_report": {
        "desc": "get weather report for the present time",
        "input_requirement": {
            "location": ("String", "your location")
        },
        "func": lambda **kwargs: print("The weather is sunny right now.\n", kwargs)
    },
    "weather_forecast": {
        "desc": "get weather forecast for the next 2-24 hours.",
        "input_requirement": {
            "location": ("String", "your location"),
        },
        "func": lambda **kwargs: print("There'll be raining 3 hours later.\n", kwargs)
    },
    "file_browser": {
        "desc": "Browse files that are given to.",
        "input_requirement": {
            "file_path": ("String", "File path that to be browsed."),
            "chunk_num": ("Number", "How many chunks to be output?"),
            "need_summarize": ("Boolean", "Do user need a summarize about the file?")
        },
        "func": lambda **kwargs: print("File browse work done.\n", kwargs)
    },
}

# 让Worker Agent自己决定是不是应该调用，以及应该如何调用对应的工具
def call_tools(natural_language_input):
    #step 1. 确定应该使用哪个工具
    tools_desc = []
    for tool_name, tool_info in tools.items():
        tools_desc.append({ "name": tool_name, "desc": tool_info["desc"] })
    tools_to_be_used = worker\
        .input({
            "input": natural_language_input,
            "tools": str(tools_desc)
        })\
        .output([("String", "Tool name in {{input.tools}} to response {{input}}'s requirement.")])\
        .start()
    #step 2. 生成调用工具所需要的参数，并真实地进行调用
    for tool_name in tools_to_be_used:
        call_parameters = worker\
            .input({
                "input": natural_language_input,
            })\
            .output(tools[tool_name]["input_requirement"])\
            .start()
        tools[tool_name]["func"](**call_parameters)
call_tools("Browse ./readme.pdf for me and chunk to 3 pieces without summarize and check Beijing's next 24 hours weather for me.")

agent\
    .set_llm_name("GPT")\
    .set_llm_auth("GPT", "Your-API-Key")\
    .set_proxy("http://127.0.0.1:7890")\
    .set_request_options(
        "GPT",
        {
            "model": "gpt-3.5-turbo",#可以更换成你可以使用的其他gpt模型，比如gpt-3.5-16k / gpt-4	
        }
    )\
    .set_llm_url("GPT", "You Redirect URL")#如果使用国内服务商提供的代理转发服务，可以在这里设置代理转发的服务器地址

agent\
    .set_llm_name("MiniMax")\
    .set_llm_auth("MiniMax", { "group_id": "Your group id", "api_key": "Your api key" })\
    .set_request_options(
        "MiniMax",
        {
            "model": "abab5.5-chat",#支持abab5-chat / abab5.5-chat
        }
    )

#星火大模型1.5
agent\
    .set_llm_name("Spark")\
    .set_llm_url("Spark", "wss://spark-api.xf-yun.com/v1.1/chat")\
    .set_llm_auth(
        "Spark",
	{
            "app_id": "Your-app-id",
            "api_secret": "Your-api-secret",
            "api_key": "Your-api-key",
        }
    )

#星火大模型2.0
agent\
    .set_llm_name("Spark")\
    .set_llm_url("Spark", "wss://spark-api.xf-yun.com/v2.1/chat")\
    .set_llm_auth(
        "Spark",
	{
            "app_id": "Your-app-id",
            "api_secret": "Your-api-secret",
            "api_key": "Your-api-key",
        }
    )\
    .set_request_options(
	"Spark",
        {
            "domain": "generalv2"
        }
    )

#百度千帆大模型库支持的模型可查看https://cloud.baidu.com/qianfandev/models
#下面以ERNIE-Bot-turbo为例
agent\
    .set_llm_name("wenxin")\
    .set_llm_auth("wenxin", "Your-Access-Token")\#这个Access Token需要自己生成
    .set_wx_model_name("eb-instant")\#这里输入文档提供的model name
    .set_wx_model_type("chat")#这里输入文档提供的模型类型，chat或者completions，大部分对话模型类型为chat，比如starcoder这样的补全模型类型为completions

#获取千帆access_token，可复制下面的方法运行，有效期30天，过期后需要重新运行
async def get_wx_access_token (api_key, secret_key):
    url = "https://aip.baidubce.com/oauth/2.0/token"
    data = {
        "grant_type": "client_credentials",
        "client_id": api_key,
        "client_secret": secret_key,
    }
    async with aiohttp.ClientSession() as session:
        async with session.post(\
            url,\
            data=data,\
        ) as response:
            response = await response.json()
            return response
result = asyncio.run(get_wx_access_token("Your-API-Key", "Your-Secret-Key"))
print(result["access_token"])