cachegpt is a lightweight package that provides a persistant cache
for (some) calls to the OpenAI API, using the diskcache package.
It is primarily intended for use in interactive data science work,
where it can save you a considerable amount of time and money.
import cachegpt
# Create client. Optional arguments (e.g. model) are passed to the OpenAI completions API
gpt = cachegpt.GPT(cache_dir='data/gpt_cache', model='gpt-3.5-turbo')
# First call
%time gpt("Hello world", system_prompt="Respond in French")
# CPU times: user 24.6 ms, sys: 22 ms, total: 46.6 ms
# Wall time: 650 ms
# 'Bonjour le monde!'
# Second call - uses cache
%time gpt("Hello world", system_prompt="Respond in French")
# CPU times: user 1.11 ms, sys: 696 µs, total: 1.81 ms
# Wall time: 1.2 ms
# 'Bonjour le monde!'embeddings = cachegpt.Embeddings(cache_dir='data/embedding_cache')
inputs = ["apple", "banana", "Mexico"]
# Default output is a dataframe with a column for each input
embeddings(inputs)
# apple banana Mexico
# 0 0.007811 -0.013927 -0.000323
# 1 -0.022955 -0.032886 -0.006600
# 2 -0.007400 0.007615 -0.001157
# 3 -0.027779 -0.016562 -0.024243
# 4 -0.004670 -0.005112 -0.007740
# ... ... ... ...
# 1531 0.021954 0.025218 0.001689
# 1532 -0.012101 -0.017800 -0.001647
# 1533 -0.013581 -0.006548 0.002300
# 1534 -0.015678 -0.017102 -0.018005
# 1535 0.006102 0.001874 -0.027697
# [1536 rows x 3 columns]By default, GPT() and Embeddings() use dotenv to look for a .env file and read in OPENAI_KEY as an environment variable.
Alternatively, you can pass an API key directly (by setting api_key=="<your-key>"), or prompt the user for one (auth=="prompt").