LLLMML

This is meant to be a markup language for interacting with LLM models, similar to Hero ML. This is written from the ground up with simplicity, minimalism, and power usage in mind.

Core features currently include:

• Variables!
  • These can be from a user supplied dictionary or special variables like: $1, previous inputs
• Control flow!
  • Extremely basic if statementst are provided
  • Check TODO for more info

Included in this repo is a python library for compiling this script and using it with your own programs. Here is a simple example script:

# You can write comments like this! All blank lines are ignored.
> Create a fictional title for the $MOVIE_ADJECTIVE movie ever! This should not be a real movie; only supply me with the title.

# blank lines and comments are all deleted
> Write a comprehensive anlaysis of $1, discussing the following details: $ANALYSIS_DETAILS. Make sure to sound as pedantic as possible.

! if 'masterpiece' in $2
> Someone approaches you and says $1 is the worst movie ever made. How do you respond?
! elif 'worst' in $2  # regex supported here!
> Some subhuman piece of garbage online actually thinks that $1 is the best movie made. How do you respond?
! else
> Someone feels passionately about $1. They are frustrated with your apathy. What do you respond with?
! endif

The script could be called like so:

temp = LLLML(template=script, llm_call=llm_call)
output = temp.compile(MOVIE_ADJECTIVE='worst', ANALYSIS_DETAILS='themes, character, and music')

# Perform sentiment analysis or something, idk. The world is your oyster!
print(output)

• LLLML is our class that wraps this functionality
• llm_call is a Callable that takes a string and returns a string from your LLM
• LLLML.compile() takes a **kwargs, named parameters

TODO:

Probably should do

• Implement a generic testing framework
  • Not sure hwo I would go about doing this but it would be nice so that the end user can simply simply testing functions

Optional

• Implement other cooler control flow
  • Again, may or may not be necessary but it might be cool
• Allow user to implement their own functions in the script
  • This may be out of the scope for the project