This package is a prototype to see how we could integrate LLM into the analyst workflow, by having it perform EDA tasks.
tidychat’s approach is to make it easy to test multiple LLM’s in the
back end,by separating the integration with a given API, from the UI of
this package.
This package also provides a flexible way to view and customize the prompts being sent to the LLM.
At this time, only the OpenAI API is supported. And within that API,
there are two major internal functions that communicate with this API.
The first one is the API “Chat Completion” which is only available for
GPT 3.5 or above. The other is for “Completion”, which is used for the
“lower” OpenAI models, such as DaVinci and Ada. Switching between the
two is as simple as changing the default of the model argument from
gpt3.5-turbo to text-davinci-003, for example. tidychat will take
care of switching between the two OpenAI endpoints (Completion or Chat
Completion) seamlessly.
Since this is a very early version of the package, you can either clone the repo, or install the package from GH:
remotes::install_github("edgararuiz/tidychat")Please refer to the video I shared on Slack to see how to use it.