s115i24, 8:41 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp Blog > Chatgpt API Exploring the ChatGPT API with Python Exploring the ChatGPT API with Python: Multiple Completions, Max Tokens, ... OpenAl recently introduced? the ChatGPT API along with the Whisper API, which has already been integrated into some popular apps such as Snapchat and Instacart. The new ChatGPT API features a new model called gpt-3.5-turbo that has replaced the previously most commonly used model, text-davinci-e03 Join the AI BootCamp! Ready to dive into the world of Al and Machine Learning? Join the Al BootCamp to transform your career with the latest skills and hands-on project experience. Learn about LLMs, ML best practices, and much more! JOIN NOW ntpshwwwmlexperoxblogichatgpt-ap wtrsa, e:1aM Explering the ChaGPT API wth Python | MLEeper - Get Things Done wth AI Botcamp This new model is the exact same model used in the ChatGPT web app. Also, itis about 10x cheaper! The OpenAl team recommends the new model as a replacement for all previously used models. This part will utilize the official Python library provided by OpenAl, which includes a new component called ChatCompletion specifically designed for ChatGPT. Let's get started! © In this tutorial, we will be using Jupyter Notebook to run the code. If you prefer to follow along, you can access the notebook here: open the notebook Get API Key To use the API, you will need to register for an account on the OpenAl platform and generate an API key from this page: https://platform.openai.com/account/api-keys Setup You'll need the openai @ library, version @.27.0 . We'll also use the tiktoken 3 to count number of tokens in a given text: pip install -qqq openai==0.27.0 pip install -qqq tiktoken==0.3.0 Next, let's add imports and set the API key: import openai import tiktoken from IPython.display import display, Markdown openai.api_key = "YOUR API KEY" https shwwn-mlexpertiafblogichatgpt-ap| ams115i24, 8:41 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp Call the API To start with the APIS, we'll need a prompt. Let's reuse the prompt from the prompt engineering guide: prompt = Write me a plan of how to invest 10,0@ USD with the goal of making maximum profit in 1 Give specific investments and allocation percentages. Next, we'll make the API call: result = openai.ChatCompletion.create( model="gpt-3.5-turbo", message: {"role": "system", “content”: "You're an expert personal finance advisor."}, {"role": "user", "content": prompt} 1. d result The model parameter specifies the version of the GPT model to use, in this case gpt- 3.5-turbo . The messages parameter is a list of two dictionaries representing the conversation between the user and the model. The first dictionary has a system role and provides a statement to set the context for the conversation. The second dictionary has a user role and contains the prompt. The API response is in JSON format and contains information regarding the response generated by ChatGPT: { "choices": [ { “#inish_reason": “stop, "index": @, "message": { ntps:hwwwmlexperofblogichatgpt-ap aits115i24, 8:41 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp "content": "\n\nBefore creating an investment plan, it\u2019s important to note assistant” 1678135181, hatcmp] -6rBuPqnXPFt fdLNOcnkvAoldGykQ3" , 1 "gpt-3.5-turbo-03e1", "object": "chat.completion”, “usage”: { “completion_tokens": 397, “prompt_tokens": 54, “total_tokens": 451 Let's have a look at two key-value pairs © choices: an array that contains an object with the response message, index, and finish_reason * usage: an object that provides the number of tokens used to generate the response In this case, the choices array contains only one object, which represents the response message generated by ChatGPT. The message contains a recommended investment plan and some advice for the user. The usage object indicates the number of tokens used to generate the response, which can be useful to track usage and costs. We will use a helper function to extract the message content and display it in Markdown format def show_choice(choice) display (Markdown (choice[ "message"]["content"])) show_choice(result["choices"][@]) [Bi chatcrt Response Conversations https shwww-mlexpertiafblogichatgpt-ap| amssi, sat aM Espoing the ChatGPT AP with Python| MLExpert- Got Things Done wt Al Booteamp The messages parameter is designed to contain the whole conversation between the user and ChatGPT. It's important to note that the API does not have the ability to record or remember previous conversations. To create a conversation with ChatGPT, we can use the result object from the previous API call and add the generated response to the messages list. Then, we can make another API call using this extended messages list to generate a tweet about the personalized finance plan: answer = result[*choices"][]["nessage”][ "content" ] format_prompt "Write the plan as a tweet” result = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ “system”, “content”: “You're an expert personal finance advisor."}, "user", “content”: prompt}, “assistant”, “content”: answer}, “user”, "content": format_prompt}, show_choice(result[ "choices" ][@]) Hi chatGrr Response Investing 60% in stocks, 20% in alternative investments, 10-20% in cash reserves, and remaining 10-20% in bonds can maximize profit in 1 year for $10K portfolio. However, always do your research and seek professional advice. #personal finance #investing #portfolioallocation API Options The API has a wide array of options’, We'll have a look at limiting the number of tokens, getting multiple completions and adjusting temperature. Maximum Tokens https shwwn-mlexpertiafblogichatgpt-ap| sits115i24, 8:41 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp ChatGPT's pricing is token-based, meaning the cost is determined by the number of tokens you use. There is a maximum token limit of 4096 per API request. Both input and output tokens count toward this limit and your payment. To control the maximum number of tokens generated by the model, the max_tokens parameter can be specified response = openai .chatCompletion.create( model="gpt-3.5~turbo" messages=[ (role ‘You're an expert personal finance advisor. (role pronpt} L max_tokens=256 ) response { "choices": [ { “finish_reason": "Length", "index": @, “message t content": "\n\nAs an AI language model, I cannot provide investment advice or assistant” 1678135229, id”: "chatempl-6rBvBDH2qjSFhv@lwPqB1az7dBTU" , model": "gpt-3.5-turbo-0301", "object": "chat.completion”, sage": { ‘completion_tokens": 256, prompt_tokens": 54, total_tokens": 310 https shwwn-mlexpertiafblogichatgpt-ap| eit9115724, 841 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp Note that the finish_reason isnow stop and completion_tokens is equal to the number of max_tokens - 256. If you want to know how many tokens your prompts contain, you can use the "tiktoken" library to create this function: def num_tokens_fron_messages (messages): Returns the number of tokens used by a list of messages. Code from: https: //github.com/openai /openai -cookbook encoding = tiktoken.encoding_for_model("gpt-3.5-turbo") num_tokens = @ for message in messages num_tokens += 4 # every message follows {role/name}\n{content}assistant return num_tokens The num_tokens_from_messages function takes a list of messages and returns the number of tokens that would be used by ChatGPT. Let's check the last messages: messages = [ » "content": "You're an expert personal finance advisor."}, content": prompt} num_tokens_from_messages (messages) 54 The number 54 is the exact same value returned by the usage.prompt_tokens attribute of the API response. This attribute indicates the number of tokens used by the prompt provided in the API request. ntps:hwwwmlexperofblogichatgpt-ap mms115i24, 8:41 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp Generate Multiple Responses To generate multiple completions for a prompt using ChatGPT API, you can set the n parameter to specify the number of completions you want result = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ “content”: "You're an expert personal finance advisor."}, prompt} for i, choice in enumerate(result["choices"]): print(f"Response {i + 1}:") show_choice(choice) print("\n") [HB chatort Response Temperature You can adjust the level of randomness and creativity in the responses generated by ChatGPT using the temperature parameter. The temperature ranges from 0 to 2, with a default value of 1. A lower temperature will result in more focused and predictable responses, while a higher temperature will result in more diverse and unpredictable responses: result = openai.chatCompletion.create( model="gpt-3.5-turbo", messages=[ "system", : "user", " "You're an expert personal finance advisor."}, : prompt} iy temperature=0 d show_choice(result["choices"][@]) https shwwn-mlexpertiafblogichatgpt-ap| amt115i24, 8:41 AM Exploring the ChatGPT API with Python | MLExpert- Get Things Done with Al Bootcamp HB chatarr Response result = openai.ChatCompletion.create( model="gpt-3.5-turbo" messages=[ {("role": "system", “content”: "You're an expert personal finance advisor."}, {role": "user", “content”: prompt} iy ‘temperature=1 ) show_choice(result, choices” [e]) HB chatert Response result = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ role "content": "You're an expert personal finance advisor. ("role content": prompt} L temperatun’ ) show_choice(result["choices"][@]) Hi chaterr Response response = openai .chatCompletion.create( gpt-3.5-turbo", system", “content user", ‘You're an expert personal finance advisor content": prompt} temperature ) show_choice(response["choices"][@]) [HB chaterr Response ntps:hwwwmlexperofblogichatgpt-ap omns, :41aM Explring the ChaGPT API wth Python | MLEeper- Get Things Done wth Al Boicamp Gradually increasing the temperature parameter can lead to more random and nonsensical completions, while finding the right balance of values can result in more focused and relevant responses Conclusion In conclusion, we have explored various features and functionalities of OpenAl's ChatGPT API. We have seen how to control the number of tokens, get multiple completions, adjust the temperature, and extract message content using helper functions. Compared to the web app, the ChatGPT API offers significantly more customization options. One particularly interesting area to explore is creating and modifying conversations, which can potentially result in better response generation. 3,000+ people already joined Join the The State of Al Newsletter Every week, receive a curated collection of cutting-edge Al developments, practical tutorials, and analysis, empowering you to stay ahead in the rapidly evolving field of AL Your Email Address SUBSCRIBE Iwon't send you any spam, ever! References 1. ChatGPT and Whisper APIs @ 2. Qpenai Python Library © https shwwn-mlexpertiafblogichatgpt-ap| somssi, st aM Espoing the ChatGPT AP with Python| MLExpert- Got Things Done wt Al Booteamp Tiktoken Library & ChatGPT API Tutorial © 5. ChatGPT API Reference © hitpsswww-mlexper ioblog/chatgpt-api awit