Project CodeNet by IBM

Project CodeNet is an Artificial Intelligence (AI) dataset created by IBM for the purpose of teaching AI how to code. It contains around 14 million code examples and around 500 million lines of code in more than 55 different programming languages, from the most modern like C++, Java, Python, and Go to the more legacy languages such as COBOL, Pascal, and FORTRAN.

AI for Code is on the verge of being widely adopted. To make this happen, researchers from IBM Research have initiated Project CodeNet, a large-scale dataset for testing and evaluation. Project CodeNet has many of the same characteristics (large scale, diversity, etc.) as ImageNet, a massive dataset for images that had a significant effect on the field of computer vision research. Project CodeNet is a large dataset composed of around 14 million code samples, each of which is an intended answer to one of 4000 coding challenges. It is hoped that Project CodeNet will have the same effect on AI for Code as ImageNet had on computer vision.

Source: https://research.ibm.com/blog/codenet-ai-for-code