Long Context Code
Long context code models aim to improve the ability of large language models (LLMs) to understand and generate code from significantly larger input contexts, such as entire software projects, rather than just individual functions. Current research focuses on scaling existing LLMs to handle vastly increased context windows (e.g., up to 128,000 tokens) through techniques like modified attention mechanisms and efficient data pre-processing. This work is driven by the need for more robust and realistic code evaluation benchmarks and the potential to improve various software engineering tasks, including bug detection, code completion, and automated code generation.
Papers
July 18, 2024
June 17, 2024
June 10, 2024
December 19, 2023
December 8, 2023