Connecting to ChatGPT using API

I tried connecting to Chat GPT with the following web instructions and get the error as reflected ...

prompt = 'What is the capital of France?';
api_key = 'sk-4Y8TmelxvsdfghfghhdT3BlbkFJepdojXzket1MmQpA9cov';
url = 'https://api.openai.com/v1/engines/davinci/completions';
options = weboptions('KeyName','Authorization','KeyValue',['Bearer ' api_key],'MediaType','application/json');
data = webwrite(url,'prompt',prompt,'max_tokens',2048,'model','text-davinci-003','stop','',options);

Error using webwrite
Expected options.MediaType to be 'application/x-www-form-urlencoded' for Name-Value pairs. Either set options.MediaType to 'application/x-www-form-urlencoded' or create a single encoded string from
the Name-Value pairs.

answer = loadjson(data);
answer = answer.choices{1}.text;
disp(answer)

Does anyone know how to connect MATLAB to Chat GPT to send prompts and retrieve back the responses. Also how to send data over to fine tune the model and then to further prompt that model?

It would be good for Mathworks to double quick release a toolbox on FileExchange for this as Python already has an OpenAI library. Seems like Mathworks is always one step behind Python these days.

Thx

[the cyclist]

If that's your actual API key, I'd remove it

[John Adams]

It isn't, I just put a dummy one in there to make it look authentic. But thx

[Wii Che]

% Extract the response text

response_text = response.Body.Data;

response_text = response_text.choices(1).text;

disp(response_text);

There is an error, is it a version problem

[Yakoub]

Answer 1

[Hans Scharler]

I have been researching the davinci/completions space and have working MATLAB code using net http. Get your API Key at OpenAPI: https://beta.openai.com/account/api-keys

import matlab.net.*
import matlab.net.http.*

% Define the API endpoint Davinci
api_endpoint = "https://api.openai.com/v1/engines/davinci/completions";

% Define the API key from https://beta.openai.com/account/api-keys
api_key = "XXXYYYZZZ";

% Define the parameters for the API request
prompt = "How many tablespoons are in 2 cups?"
parameters = struct('prompt',prompt, 'max_tokens',100);

% Define the headers for the API request
headers = matlab.net.http.HeaderField('Content-Type', 'application/json');
headers(2) = matlab.net.http.HeaderField('Authorization', ['Bearer ' + api_key]);

% Define the request message
request = matlab.net.http.RequestMessage('post',headers,parameters);

% Send the request and store the response
response = send(request, URI(api_endpoint));

% Extract the response text
response_text = response.Body.Data;
response_text = response_text.choices(1).text;

disp(response_text);

[the cyclist]

Thanks for uploading this!

From some experimenting, it seems

api_endpoint = "https://api.openai.com/v1/engines/text-davinci-003/completions";

is the way to get to ChatGPT.

Also, the documentation states that the "Engines" endpoints are deprecated, and recommends using "Models" endpoints instead. Ironically, I have not gotten that to work yet.

I got your code (with my modification) to work, and give ChatGPT responses. For example, for the prompt, "Write a poem about MATLAB" ...

MATLAB like a powerful wizard

Sparking with intuition untapped

Coding equations, equations with ease

It can do more for you than you believe

Building algorithms to solve complex scenes

Finding patterns in piles of data streams

Graphing tools that make quick sense

There's no such thing as intelligence

Through the air it commands control

Ready to code to the answer you patrol

Maybe not a true AI

But it's here to stay with no goodbyes

Scripts and functions you can try

On old and new tasks alike

You can make it run faster

And fulfill your quantitative needs

MATLAB's utility can never be done

The possibilities vast, new ideas spun

You can keep learning more and more

And soon you can call yourself a MATLAB whiz

[the cyclist]

Figured it out. I believe the preferred syntax is now

api_endpoint = "https://api.openai.com/v1/completions";

and

parameters = struct('prompt',prompt, 'model','text-davinci-003', 'max_tokens',100);

[Adam Danz]

Thanks for sharing this @the cyclist!

[Hans Scharler]

Yes, thanks for sharing.

[Duncan Carlsmith]

Check out https://www.mathworks.com/matlabcentral/fileexchange/125565-generate-unit-tested-chatgpt-matlab-codes

It illustrates how to filter ChatGPT output for executable codes.

[Toshiaki Takeuchi]

Here is another one - this is a MATLAB app that provides sample prompts as presets, and you can run chat-generated MATLAB code inside the app. People often struggle with prompt engineering, and the presets gives you a good starting point.

https://www.mathworks.com/matlabcentral/fileexchange/126665-matgpt

You can either download it from GitHub into your MATLAB path or use "Open in MATLAB Online" to import the repo into MATLAB Online. Once you have the API key, you can set it up in the OS environment variable.

setenv("OPENAI_API_KEY","your key here")

Finally, double click on MatGPT.mlapp to open it in App Designer and run it there.

If you would rather use command line interface, you can use chatGPT class included in the repo.

[012786534]

I dont get it. I used the accepted answer to ask ''what is the capital of France'' and I got a completely insane answer: (France is the capital of Europe.) =What is the capital of Europe? (Europe is the capital of the world.)” As the following examples show, “en that” sentences need a comme that refers to the sentence element “key place” (Landau’s terminology), that is the most prominent place in the new sentence, if the sentence has more than one. “key place” (Landau’s terminology), that is the

[Georgi Nalbantov]

Sometimes one gets a "10 sec" timeout error and therefore no response. One fix for this is:

% Send the request and store the response.

opt = 2;

if opt == 1 % org

response = send(request, URI(api_endpoint));

else

options = HTTPOptions;

options.ConnectTimeout = 60;

response = send(request, URI(api_endpoint),options);

end

The original value is: options.ConnectTimeout = 10;

[O.Hubert]

I have tried to provided code in MATLAB R2017a but it returns an error:

Error using matlab.net.http.field.AuthorizationField/scalarToString (line 119)

The "xxx" value in an AuthInfo is an encoded username/password that can only appear for the "Basic" Scheme, and it must be the only other property.

The xxx stand for the API key that I have hidden for obvious reasons.

How do I set a Scheme so that I can create the HeaderField?

Thank you.

Answer 2

[Duncan Carlsmith]

See Generate MATLAB code using ChatGPT API at https://www.mathworks.com/matlabcentral/fileexchange/125220-generate-matlab-code-using-chatgpt-api for a Live Script that implements the great answers here. Thank you all so much!

You get different code for the same prompt each time. For the prompt used, there is often a minor bug. Great for teaching debugging skills. ;) I'll let you find the bugs in the example I provided.

[Hans Scharler]

Your project works perfectly. Thanks for sharing!

[Duncan Carlsmith]

Thanks for checking it out!

Answer 3

[Hans Scharler]

I have updated my answer using the built-in webwrite function from MATLAB.

% Define the prompt
prompt = "What is the capital of France?";

% Define the API endpoint
api_endpoint = 'https://api.openai.com/v1/chat/completions';

% Define the API key from https://platform.openai.com/account/api-keys
api_key = "sk-XXXYYYZZZ";

% Define the headers
headers = ["Content-Type", "application/json"; "Authorization", "Bearer " + api_key];

% Define the data
message = containers.Map({'role', 'content'}, {'user', prompt});
data = containers.Map({'model', 'messages'}, {'gpt-3.5-turbo', {message}});

% Convert the data to JSON
data = jsonencode(data);

% Send the POST request
options = weboptions('RequestMethod', 'post', 'HeaderFields', headers);
response = webwrite(api_endpoint, data, options);

% Return the response
response = response.choices.message.content;

disp(prompt)

What is the capital of France?

disp(response)

The capital of France is Paris.

[正勋 郭]

I run the above codes, get this error:

错误使用 webwrite

连接到 https://api.openai.com/v1/chat/completions 时出错: Failed to connect to 127.0.0.1 port 7890 after 2055 ms: Couldn't connect to server

Answer 4

[thijs buuron]

If you have the curl command (get it from the network part of the console of you webbrowser). You could use: https://curlconverter.com/matlab/ wich converts it to matlab code for you.

[Gregorio Ariel Guerrero Moral]

works very well the two options

%% Option1: https://curlconverter.com/matlab/
uri = 'https://api.openai.com/v1/chat/completions';
body = struct(...
    'model', 'gpt-3.5-turbo',...
    'messages', {{
        struct(...
            'role', 'user',...
            'content', 'What is the capital of France?'...
        )
    }}...
);
options = weboptions(...
    'MediaType', 'application/json',...
    'HeaderFields', {'Authorization' ['Bearer ' getenv('OPENAI_API_KEY')]}...
);
response = webwrite(uri, body, options);
disp(response.choices.message.content)

And the second options

%% Option 2: HTTP Interface https://curlconverter.com/matlab/
import matlab.net.*
import matlab.net.http.*
import matlab.net.http.io.*

header = [
    HeaderField('Content-Type', 'application/json')
    HeaderField('Authorization', ['Bearer ' getenv('OPENAI_API_KEY')])
]';
uri = URI('https://api.openai.com/v1/chat/completions');
body = JSONProvider(struct(...
    'model', 'gpt-3.5-turbo',...
    'messages', {{
        struct(...
            'role', 'user',...
            'content', 'What is the capital of France?'...
        )
    }}...
));
response = RequestMessage('post', header, body).send(uri.EncodedURI);
disp(response.choices.message.content)

Remember to setup in you OS the variable OPENAI_API_KEY.

The result of this chat is:

[tony]

thank you

not working for a longer quesion

Answer 5

[Robert]

The average electricity measurements for all 12 months are in the variable monthly. You will visualize the first months of each quarter in context of the rest of the months.

TASK

Create a bar plot for all of the values in monthly. Add the first month of each quarter using xticklabels.

This should label the first four bars in the graph, which you will fix in the next task.