Documentation for Main Files

This is the main file used to run MARCO. It contains the specifications for MARCO’s UI, and is also where the functions used to optimize code are called.

Lines 1-5 import the required packages and files. streamlit is the framework used to build MARCO’s frontend. agent_functions contains the functions that concern interacting with the LLM. asyncio is a Python library that allows for concurrent programming, allowing process_optimization() to wait for input from functions in agent_functions. base64 was used for encoding a background image and is no longer used in the file. time is used in the typewriter_effect_code function to simulate dynamic text output.

Lines 8-83 set the formatting for MARCO’s webpage using the markdown() function from streamlit.

Lines 85-92 define create_error_message(), which creates error messages as needed.

Lines 95-102 define typewriter_effect_code(), which allows the text to output as if it is being typed one letter at a time, instead of all at once.

Lines 105-106 initialize the session.

Lines 109-114 control the layout of the main content area (i.e., the title, the textbox for the code, and the controls for selecting the LLM and number of techniques used)

Lines 116-144 define the process_optimization(), which gets the optimization techniques/optimized code using functions in agent_functions.py and outputs them (along with the unoptimized code imputed by the user, for comparison purposes.

Lines 146-150 govern what happens when the user presses the “Improve Code” button. If there is code in the textbox, process_optimization() is called to start the optimization process. Otherwise, a warning message is outputted asking the user to input code.

This file contains the functions that require direct interaction with the LLM. These functions are called within process_optimization() in agent.py.

Lines 1-7 import the required packages. The primary framework used is LangChain, allowing us to power MARCO with models developed by OpenAI. Most of the other imports deal with file access.

Line 9 gets necessary variables from the .env environment file, such as the API keys used for accessing the model.

Line 10 initializes llm, the variable that will be used the LLM.

Lines 13-20 define the select_model() function, which takes the model selected by the user and uses it to set llm. This function also sets llm as a global variable.

Lines 22-45 define the get_optimization_techniques() function. This function asks the model for a list of optimization techniques (with the exact number of techniques set by the user), and does some formatting before returning the list of techniques.

Lines 47-58 define the get_subsets() function. This was previously used in formatting the list of optimization techniques, but is now unused.

Lines 60-77 define the improve_code() function. This function asks the model to optimize the code using the previously generated optimization techniques, and returns the user’s imputed code optimized using each technique.