Openai_proxy

A proxy for OpenAI requests. Primarily for AI students at UPenn to track their spending on GPT-3 requests. Available for anyone to get the estimate and actual cost of a request

Price estimates in this package are based on the OpenAI API docs and pricing page They are also designed to estimate the highest possible cost of requests before they are sent. Actual request costs will always, assuming no bugs, be lower than the estimates.

A frontend GUI is available here

Installation

Install the python package using pip

pip install --upgrade openai_proxy

Setup

Setup as follows in your development environment or server (not frontend, otherwise you may get a "CORS Error")

import openai_proxy

openai_proxy.username = "koawood"
openai_proxy.course_id = "8000"
openai_proxy.access_key  = "O6rbLM3-yYcwS5GaWZEghA"
openai_proxy.access_token = "4q-_apNm72sGENVtYtcCXAnC7jEGnIUZfylHbEAEdFQ"

Check Completion Price

This request accepts the following parameters corresponding to the OpenAI API's request parameters: See the OpenAI API docs for more details
Note that this request does not require the credentials in the setup.

• prompt: the prompt to be sent to OpenAI's GPT-3 [required]
• engine: the GPT-3 engine to be used for the request
• max_tokens: the maximum number of tokens to be generated in the completion

Example:

response = openai_proxy.Completion.price(
    engine="davinci",
    prompt="What can you say about the Aztec Empire. Thousands of years",
    max_tokens=200,
    n=2,
)

response['price']

Returns a response object with a price field. Note that the Embedding and ChatCompletion classes also have price functions to estimate without sending a request.

Send Completion Request to OpenAI

This request accepts the following parameters corresponding to the OpenAI API's request parameters. See the OpenAI API docs for more details
Note that this request requires the credentials in the setup.

• prompt: the prompt to be sent to OpenAI's GPT-3 [required]
• engine: the GPT-3 engine to be used for the request
• max_tokens: the maximum number of tokens to be generated in the completion
• temperature: controls randomness. Lower temperature, lower randomness
• top_p: controls diversity. 1.0 means no diversity, 0.0 means maximum diversity
• frequency_penalty: controls frequency of words in the completion. high penalty, less repetition
• presence_penalty: controls probability of words in the completion. high penalty, newer topics
• stop: a list of tokens to stop the completion on
• n: the number of completions to generate
• best_of: the number of completions to generate and evaluate server side before returning only the best

Example:

response = openai_proxy.Completion.create(
    engine="davinci",
    prompt="What can you say about the Aztec Empire. Thousands of years",
    temperature=0.7,
    max_tokens=200,
    top_p=1,
    frequency_penalty=0,
    presence_penalty=0,
    stop=["###", '\n'],
    n=2
)

response['choices'][0]['text']
response['price']

Returns the Openai response object along with a price field.

Send Embedding Request to OpenAI

This request accepts the following parameters corresponding to the OpenAI API's request parameters. See the OpenAI API docs for more details
Note that this request requires the credentials in the setup.

• phrases: the list of phrases to be sent to OpenAI's embedding model [required]
• engine: defaults to text-embedding-ada-002 (the cheapest model)

Example:

response = openai_proxy.Embedding.create(
        phrases=["What can you say about the Aztec Empire. Thousands of years"]
    )
    
print(response)
print(response['price'])

Returns the Openai response object along with a price field.

Send ChatCompletion Request to OpenAI

This request accepts the following parameters corresponding to the OpenAI API's request parameters. See the OpenAI API docs for more details
Note that this request requires the credentials in the setup.

• messages: the list of phrases to be sent to OpenAI's embedding model [required]
• engine: defaults to gpt-3.5-turbo (the cheapest model)

Example:

response = openai_proxy.ChatCompletion.create(
        messages=[
            {
                "role": "system",
                "content": "You are a polite but sarcastic chat bot"
            },
            {
                "role": "user",
                "content": "What can you say about the Aztec Empire. Thousands of years"
            }]
    )
    print(response)
    print(response['price'])

Returns the Openai response object along with a price field.

Note that the Embedding and ChatCompletion classes also have price functions to estimate without sending a request.

Please reach out with any questions or suggestions
Ayotomiwa, tomiwa@wharton.upenn.edu