Daniel Liden

Home / Blog / About Me / Photos

Using the ChatGPT API with Julia Part 1: the HTTP.jl Library

Introduction

This brief post shows the basics of using the Julia HTTP library to interact with the OpenAI ChatGPT API, which was made public a few days ago. This post will only include the minimum necessary detail for getting started with the API. Future posts will go into a little more detail on how to send message histories and engage more interactively with the API.

curl Example

Here is the API documentation on OpenAI's website. An example request with curl looks like this:

curl https://api.openai.com/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -d '{
  "model": "gpt-3.5-turbo",
  "messages": [{"role": "user", "content": "Hello!"}]
}'

Julia example

And here's how we can re-create this with the Julia HTTP library:

OPENAI_API_KEY = ENV["OPENAI_API_KEY"]

using HTTP
using JSON

headers = HTTP.Headers([
    "Authorization" => "Bearer $OPENAI_API_KEY",
    "Content-Type" => "application/json",
])

body = json(Dict("model" => "gpt-3.5-turbo",
                 "messages" => [Dict("role" => "user", "content" => "Hello!")]))


response = HTTP.post(
    "https://api.openai.com/v1/chat/completions",
    headers,
    body;
    verbose = false,
)

# Parse the response body as JSON
result = JSON.parse(String(response.body))
print(result)
Dict{String, Any}("choices" => Any[Dict{String, Any}("finish_reason" => "stop", "message" => Dict{String, Any}("role" => "assistant", "content" => "\n\nHello there! How may I be of assistance?"), "index" => 0)], "model" => "gpt-3.5-turbo-0301", "usage" => Dict{String, Any}("completion_tokens" => 12, "total_tokens" => 21, "prompt_tokens" => 9), "id" => "chatcmpl-6qMM6OmdVRZ8VtdKgKFtb7aRDYWkF", "object" => "chat.completion", "created" => 1677937010)

Note that OPENAI_API_KEY is stored as an environment variable on my system, so I was able to access it with OPENAI_API_KEY = ENV["OPENAI_API_KEY"].

We can extract the completion itself from the "choices" dictionary entry.

chat_response = result["choices"]
chat_response[1]["message"]["content"]
"\n\nHello there! How may I be of assistance?"

A little more detail about sending messages

One of the things that makes ChatGPT useful is its ability to "remember" past parts of a conversation. We can send whole conversations to the API using the "messages" part of the request body. messages is an array of ~Dict~s, each of which has a "role" and a "content." There are three options for "role":

  1. system: sets the behavior of the assistant. The content might be, for example, you are a helpful assistant or you are a very polite customer support agent or you are a senior software engineer in a mentorship role.
  2. user: The human interacting with ChatGPT.
  3. assistant: the responses from ChatGPT

So we can send a more complete conversation and get some richer details in response. For example:

messages=[Dict("role" => "system", "content" => "You are a knowledgable and helpful Julia developer."),
         Dict("role" => "user", "content" => "Can you show me how to make a POST request with the HTTP library?")]

body = json(Dict("model" => "gpt-3.5-turbo",
                 "messages" => messages))

response = HTTP.post(
    "https://api.openai.com/v1/chat/completions",
    headers,
    body;
    verbose = false,
)

# Parse the response body as JSON
result = JSON.parse(String(response.body))
print(result["choices"][1]["message"]["content"])

```julia
using HTTP

# URL to POST to
url = "https://httpbin.org/post"

# Data to include in the POST request (in JSON format)
data = Dict("name" => "John", "age" => 30)
json_data = JSON.json(data)

# Headers to specify that we're sending JSON data
headers = Dict("Content-Type" => "application/json")

# Make the POST request
response = HTTP.request("POST", url, headers, json_data)

# Get the response body as a string
body = String(response.body)

# Print the response status code and body
println("Status code: $(response.status)")
println("Response body: $body")
```

In this example, we're sending a JSON object with a name and age property to https://httpbin.org/post, which is an HTTP testing service. The `headers` argument specifies that we're sending JSON data, while the `json_data` argument is the actual data we want to send.

The `HTTP.request` function is called with the POST method, the URL to POST to, the headers we want to send, and the data we want to include. The response is then captured in the `response` variable.

Finally, we extract the body of the response as a string using `String(response.body)`, and print both the status code and response body to the console.Yes, I can. Here's an example code snippet that shows how to make a POST request using the HTTP library in Julia:

```julia
using HTTP

url = "https://jsonplaceholder.typicode.com/posts"
data = "{\"title\":\"foo\",\"body\":\"bar\",\"userId\":1}"

response = HTTP.post(url, data, ["Content-Type" => "application/json"])
println(String(response.body))
```

In this example, we first specify the url of the endpoint we want to send our request to. Next, we create a string representation of the JSON data we want to send in our request. We use `HTTP.post()` to send a POST request to the specified url, including the JSON data in the request body, and with a content type header that specifies that the data is JSON (application/json). Finally, we print the response contents converted to a string by the `String()` function.

And if we want to send a followup referring to an earlier part in the conversation, we can extend the messages array as follows:

# include the assistant's previous response
push!(messages, result["choices"][1]["message"])

# ask a new question referring to an earlier part of the conversation
push!(messages, Dict("role" => "user",
                     "content" => "Can you please provide a shorter, simpler example?"))

4-element Vector{Dict{String, String}}:
 Dict("role" => "system", "content" => "You are a knowledgable and helpful Julia developer.")
 Dict("role" => "user", "content" => "Can you show me how to make a POST request with the HTTP library?")
 Dict("role" => "assistant", "content" => "Yes, I can. Here's an example code snippet that shows how to make a POST request using the HTTP library in Julia:\n\n```julia\nusing HTTP\n\nurl = \"https://jsonplaceholder.typicode.com/posts\"\ndata = \"{\\\"title\\\":\\\"foo\\\",\\\"body\\\":\\\"bar\\\",\\\"userId\\\":1}\"\n\nresponse = HTTP.post(url, data, [\"Content-Type\" => \"application/json\"])\nprintln(String(response.body))\n```\n\nIn this example, we first specify the url of the endpoint we want to send our request to. Next, we create a string representation of the JSON data we want to send in our request. We use `HTTP.post()` to send a POST request to the specified url, including the JSON data in the request body, and with a content type header that specifies that the data is JSON (application/json). Finally, we print the response contents converted to a string by the `String()` function.")
 Dict("role" => "user", "content" => "Can you please provide a shorter, simpler example?")

And then request another response:

body = json(Dict("model" => "gpt-3.5-turbo",
                 "messages" => messages))

response = HTTP.post(
    "https://api.openai.com/v1/chat/completions",
    headers,
    body;
    verbose = false,
)

# Parse the response body as JSON
result = JSON.parse(String(response.body))
print(result["choices"][1]["message"]["content"])
Sure, here's a simpler example:

```julia
using HTTP

url = "https://jsonplaceholder.typicode.com/posts"

response = HTTP.post(url, form = [("title", "foo"), ("body", "bar"), ("userId", "1")])
println(String(response.body))
```

In this example, we are sending a POST request with form data rather than JSON data. We specify the form data using an array of tuples with the keys and values for the form fields. Note that we use the `form` argument to pass the form data to `HTTP.post()`. The response is then printed in the same way as before.

Next Up…

This post showed basic usage of the ChatGPT API with Julia. In the next post, I'll show how to make this more modular and useful. We'll create a Struct for conversations and a function to call on the API based on the conversation history in that Struct. After that, I'll write about how to make it interactive, perhaps as a Julia REPL mode, but at least as a command line utility.

Date: 2023-03-04 Sat 00:00

Emacs 27.1 (Org mode 9.3)