Evaluating Large Language Models Trained on Code

TL;DR

This paper introduces Codex, a GPT-based model fine-tuned on GitHub code, demonstrating significant improvements in Python code synthesis and exploring strategies like sampling to enhance problem-solving success.

Contribution

The paper presents Codex, a new large language model trained on code, and evaluates its capabilities, including a novel benchmark and insights into its strengths and limitations.

Findings

Codex solves 28.8% of HumanEval problems, outperforming GPT-3 and GPT-J.

Repeated sampling significantly increases solution success rate.

Identifies limitations in handling long chains of operations and variable binding.

Abstract

We introduce Codex, a GPT language model fine-tuned on publicly available code from GitHub, and study its Python code-writing capabilities. A distinct production version of Codex powers GitHub Copilot. On HumanEval, a new evaluation set we release to measure functional correctness for synthesizing programs from docstrings, our model solves 28.8% of the problems, while GPT-3 solves 0% and GPT-J solves 11.4%. Furthermore, we find that repeated sampling from the model is a surprisingly effective strategy for producing working solutions to difficult prompts. Using this method, we solve 70.2% of our problems with 100 samples per problem. Careful investigation of our model reveals its limitations, including difficulty with docstrings describing long chains of operations and with binding operations to variables. Finally, we discuss the potential broader impacts of deploying powerful code generation technologies, covering safety, security, and economics.