Steer is a high-level pseudocode language designed to guide LLMs in performing tasks based on user input. It is specifically designed for users to interact with language models and instruct them to perform operations.
The grammar includes basic control flow operators, for / while loops, variable and function declaration.
It also includes some functionalities designed for LLM interaction:
ai: explain what you want to do in a sentence and pass inputsextract: used to extract information from the inputwrite: used to include strings in the final output of the model
Tip: it's best to use write as you would log something. This makes the model "think out loud", improving the accuracy of the results.
function gcd(a, b):
return ai(`find the greatest common divisor between {a} and {b}`)
function main(input_data):
print("Extracting information from the input data")
a = extract(input_data, 'a')
b = extract(input_data, 'b')
print("Computing the gcd of a and b")
result = gcd(a, b)
print(`The gcd is {a}`)
- Create a python venv (
python3 -m venv venv) and install the deps (pip3 install -r requirements.txt) - Set your
OPENAI_API_KEYas an env variable.
To generate the lexer, parser, and listener files from the grammar, run the following command:
antlr4 -Dlanguage=Python3 antlr/steer.g4 -o src/
You can now create your .steer script (see examples folder for examples).
Look at run_example.py to see how to use the steer_utils to run your steer file.
You can run the example with: python3 src/run_example.py
You should expect to see this:
Validating steer script with ANTLR
Parsing completed successfully!
Model output:
Extracting information from the input data
Computing the gcd of a and b
The gcd is 6
The project is structured as follows:
antlr/: Contains the ANTLR grammar file for Steer language (steer.g4).examples/: Contains example Steer programs.src/: Contains the source code for parsing and running Steer programs, along with any utility or helper files.
Contributions are welcomed, feel free to open issues and create PRs.