A simple test for large language models and their recall on bible verses
The original source of kjvdat.txt was lost, if found I will link it here
TODO: Switch to https://github.com/aruljohn/Bible-kjv
- Install golang https://golang.org/
- Open a terminal, clone this project, set it to active directory
- Update packages with
go mod tidy
- Run the program with
go run main.go
- Configure the code in the function
getLLM()
to change the URL and authentication token of your LLM provider. - Change the comments on the
BookNames
variable to change which books to run the test on. This is useful as the test may take some time to complete.
The program will output the overall accuracy, the percentage of verses that the language model correctly recalled.
It will also print CSV text that can be easily imported into excel and visualized with conditional cells.
See this example excel formatting of a test with llama3 on 2 Peter (11% accuracy)
Contributions are welcome, please keep the code as simple as possible as this is meant to be a very quick and easy test to manage.