Several mini-projects exploring the OpenAI API exploring ideas from the OpenAI bootcamp ChatGPT, GPT4, DALLE, etc.
response = openai.Moderation.create(input=”Sample text goes here”)
output = response[“results][0]Endpoint to test whether a prompt violates moderation policies
GPT-3 (Davinci)
- $0.02 / 1k tokens (approx 750 words)
- Introduction of ChatGPT 3.5 and Whisper API
GPT-3.5 (gpt-3.5-turbo)
- $0.002 / 1K tokens
- As of 3/1/2023 fine-tuning is not available for the gpt-3.5-turbo models
Whisper (transcription of audio) is now available as an API
- $.006 / minute (rounded to the nearest second)
DALLE-2 (1024x1024 resolution)
- $0.02 / image
Training
- $0.03 / 1k tokens
Usage (Inference)
- $0.12 / 1k tokens
- Ada (Fastest) $0.0004/1K tokens
- Babbage $0.0005/1K tokens
- Curie $0.002/1K tokens
- Davinci (Most Powerful) $0.02/1K tokens
- Davinci has code variants: code-davinci-002
- Model (davinci)
- Prompt
- Temperature
- Higher values mean the model will take more risks
- Try 0.9 for more creative application,
- Try 0 (argmax sampling) for ones with a well-defined answer
- In the case of simpler tasks with a clear correct answer, lean more toward 0
- Max Tokens
- The max number of tokens to generate in the completion
- The token count of your prompt plus max_tokens cannot exceed the model’s context length
- Most models have a context length of 2048, the newest support 4096
- Top P
- An alternative to sampling with temperature called nucleus sampling
- The model considers the results of the tokens with top_p probability mass:
- 0.1 means only consider the tokens comprising the top 10% probability mass.
- Use Top P or Temperature but not both
- N
- How many completions to generate for each prompt
- This is the same as running the same prompt multiple times, so it is more expensive
- We will pretty much always set this to 1 indicating we want 1 completion
- This should only be used if the temp is set above zero
- Multiple completions with a 0 Temp should generate the same result
- Frequency Penalty
- Number between -2.0 and 2.0
- Positive values penalize new tokens based on their existing frequency in the text
- Positive values reduce the likelihood of the completion repeating itself
- Presence Penalty
- Number between -2.0 and 2.0
- Positive values penalize new tokens based on whether they appear in the text so far
- This will increase the mode’s likelihood to talk about new topics
- High numbers of Freq/Pres can lead the completion to go off track
Presence and Frequency Penalties: The presence penalty is a one-off additive contribution that applies to all tokens that have been sampled at least once and the Frequency penalty is a contribution that is proportional to how often a particular token has already been sampled. Reasonable values of either are between 0.1 and 1 if the aim is to just reduce repetitive samples somewhat. Negative values will increase the likelihood of repetition.
-
Model:
the latest will give the best results -
Instructions:
Put instructions at the beginning of a prompt and use ### or “”” to separate out instructions from the desired output
Summarize the text below into bullet points.
Text: “””
{some text}
“””-
Detail:
instead of write a poen about OpenAI, “Write a poem about OpenAI in the style of a Haiku, make sure to only write 3 lines and focus on GPT. -
Examples:
Extract the first and last names from the text below:
Desired format:
Last Name, First Name
{some text}-
Be Direct:
Create a 3 to 5-sentence description about the product in the text below: {some text} -
Initiate the Response:
Instead of: Write a python function to scrape the wikipedia page on the world’s tallest buildings.
Try: The python function below scrapes Wikipedia to return a CSV of the world’s tallest buildings and their stats.
def scrape_wiki(
or start with import