#ACM Research Lab

ACM Research Lab

We are a student run organization that strives to create a supportive and inclusive community for people of all backgrounds and skill levels who are passionate about computer science to learn and network.

#Motivation

The motivation behind this project is to explore and implement the REDCODER model for specific tasks. REDCODER is a code generation model that likely aims to improve code-related tasks. This project revolves around implementing and understanding the REDCODER model, which is a code generation model. The team aims to evaluate the effectiveness of REDCODER for specific tasks by comparing its performance under different scenarios: converting code directly to a latent space versus converting it first to English and then to a latent space. Additionally, tasks include running the REDCODER codebase, understanding the associated paper, and finding a code-to-English translation model. The ultimate goal is to gain insights into the capabilities and limitations of REDCODER and potentially enhance code-related tasks through its utilization.

#Tasks

#Implement REDCODER Model:

• I am responsible for working on the Nautilus project and running the REDCODER codebase available on GitHuB, tasked with understanding the REDCODER paper.
• Additionally, there is a task to find a code-to-English codebase.

#Comparison Plan:

• Develop a plan to compare the performance of a RAG (presumably a model) under two scenarios:
  • First scenario: The code is converted to English and then to latent space.
  • Second scenario: The code is directly converted to latent space.
• Identify metrics for comparison.

#Goal

The overarching goal of the project is to evaluate the effectiveness of the REDCODER model in specific tasks. This evaluation involves comparing the performance of a model under different conversion scenarios: converting code to English first and then to latent space versus direct conversion to latent space. By achieving this goal, the group aims to gain insights into the strengths and weaknesses of the REDCODER model and potentially improve code-related tasks.