直接看一个例子吧：
| 模型 | llama3 |
prompt （核心描述） | 你是一个智能助手，可以通过拆解任务来解决问题，建议你通过执行Python代码来实现任务，如果代码成功执行后任务仍未解决，分析问题，重新审视你的假设，收集你需要的额外信息，并考虑尝试不同的方法。当你找到答案时，仔细验证答案。如果可能的话，在你的回复中包含可验证的证据。当一切都完成后，回复结束 |
| 任务 | 绘制NVDA和特斯拉股价本年迄今变化图。 将绘图保存到名为plot.png的文件中 |
| 角色 | assistant：大模型助手； user_proxy：用户代理 |
接下来先通过user_proxy对assistant发起提问，然后开始整个任务的执行：
首先user_proxy发起提问：
user_proxy (to assistant):绘制NVDA和特斯拉股价本年迄今变化图。将绘图保存到名为plot.png的文件中
assistant回复：
assistant (to user_proxy):To achieve this task, we'll need to collect the stock price data for NVDA (NVIDIA) and TSLA (Tesla) for this year and then use a Python script to plot the data.First, let's collect the data using the `yfinance` library. Here's the code to get the data:```python# filename: get_stock_data.pyimport yfinance as yfstart_date = '2023-01-01'end_date = '2023-12-31'nvda_data = yf.download('NVDA', start=start_date, end=end_date)tsla_data = yf.download('TSLA', start=start_date, end=end_date)print(nvda_data.head())print(tsla_data.head())```Please execute this code to get the stock price data for NVDA and TSLA for this year. The `print` statements will output the first few rows of the data.Once you've executed the code, please provide the output, and I'll proceed to create the plot.(Note: If you don't have the `yfinance` library installed, you can install it using `pip install yfinance`)
然后assistant开始执行：
Provide feedback to assistant. Press enter to skip and use auto-reply, or type 'exit' to end the conversation:>>>>>>>> NO HUMAN INPUT RECEIVED.>>>>>>>> USING AUTO REPLY...>>>>>>>> EXECUTING CODE BLOCK (inferred language is python)...user_proxy (to assistant):exitcode: 1 (execution failed)Code output: Traceback (most recent call last):File "/Users/lvlin/code/agent_test/demo/autogen_demo/coding_local/get_stock_data.py", line 2, in <module>import yfinance as yfModuleNotFoundError: No module named 'yfinance'
这里比较厉害了，assistant发现我环境没有yfinance，程序并没有终止，这时assistant再次向user_proxy交流，说明你环境缺几个包需要安装一下：
assistant (to user_proxy):It looks like the `yfinance` library is not installed. Let's fix that!Here's the updated code to install the `yfinance` library and then collect the stock price data:```python# filename: get_stock_data.pyimport os# Install yfinance library if not already installedos.system('pip install yfinance')import yfinance as yfstart_date = '2023-01-01'end_date = '2023-12-31'nvda_data = yf.download('NVDA', start=start_date, end=end_date)tsla_data = yf.download('TSLA', start=start_date, end=end_date)print(nvda_data.head())print(tsla_data.head())```Please execute this code to install the `yfinance` library and get the stock price data for NVDA and TSLA for this year.(Note: If you're using a virtual environment, make sure to activate it before running the code)
当然这里有一些小问题，比如模型把时间搞错了，我让它绘制的是今年的趋势，它识别成2023了，这个可以先忽略，我们可以在prompt告诉它今天的时间，模型就可以正确识别时间了；
然后assistant继续执行
Collecting yfinanceUsing cached yfinance-0.2.40-py2.py3-none-any.whl.metadata (11 kB) XXXX开始安装缺的包
安装完以后，assistant再次完善代码，加上了绘图相关代码
assistant (to user_proxy):Now that we have the stock price data, let's create the plot.Here's the updated code to plot the data:```python# filename: plot_stock_prices.pyimport yfinance as yfimport matplotlib.pyplot as pltstart_date = '2023-01-01'end_date = '2023-12-31'nvda_data = yf.download('NVDA', start=start_date, end=end_date)tsla_data = yf.download('TSLA', start=start_date, end=end_date)plt.figure(figsize=(12, 6))plt.plot(nvda_data.index, nvda_data['Close'], label='NVDA')plt.plot(tsla_data.index, tsla_data['Close'], label='TSLA')plt.xlabel('Date')plt.ylabel('Stock Price (USD)')plt.title('NVDA and TSLA Stock Prices')plt.legend()plt.savefig('plot.png')plt.show()
然后再次执行，yf.download超时了，assistant进行了代码优化
assistant (to user_proxy):It looks like the execution timed out. Let's try to optimize the code to make it more efficient.Here's the updated code:```python# filename: plot_stock_prices.pyimport yfinance as yfimport matplotlib.pyplot as pltstart_date = '2023-01-01'end_date = '2023-12-31'nvda_data = yf.download('NVDA', start=start_date, end=end_date, progress=False)tsla_data = yf.download('TSLA', start=start_date, end=end_date, progress=False)plt.figure(figsize=(12, 6))plt.plot(nvda_data.index, nvda_data['Close'], label='NVDA')plt.plot(tsla_data.index, tsla_data['Close'], label='TSLA')plt.xlabel('Date')plt.ylabel('Stock Price (USD)')plt.title('NVDA and TSLA Stock Prices')plt.legend()plt.savefig('plot.png')plt.close()# Close the figure to free up memory```I added the `progress=False` parameter to `yf.download()` to disable the progress bar, which can slow down the execution. I also added `plt.close()` to close the figure after saving it to free up memory.Please execute this code again to generate the plot and save it to a file named `plot.png`.If you still encounter issues, please let me know!
最后在反复调整之后，最终画出了股票趋势图
整个过程我们可以设置是否人为介入，如果设置为NEVER，就是模型自己反复在调整自己的策略，规划、反思、执行；如果设置了人为介入，可以在模型需要我们反馈时进行矫正或者提一些建议思路，模型都会听取进行改正。
虽然例子很简单，这个场景的代码人工写可能比模型快很多，但还是挺震撼的，整个过程没有人为干涉全部是模型一点一点自己实现的，让我们看到了大模型agent的一些潜力，未来可期
后面有机会再和大家分享一下agent在实际场景中的一些使用情况
