AI code generation

StableCode is the first programming-oriented generative AI product released by Stable AI. It uses three different models to help developers improve programming efficiency. The base model was first trained on BigCode’s stack-dataset (v1.2) and further trained on popular programming languages ​​such as Python, Go, Java, Javascript, C, markdown, and C++. In total, we trained on 560B code tokens on a high-performance computing cluster. Subsequently, by tuning the basic model, approximately 120,000 code instruction/response pairs were trained to solve complex programming tasks. StableCode is an ideal building block for learning programming, and the long text environment window model provides users with single-line and multi-line autocomplete suggestions. The model can process more code at once (2-4x more code than previously released open source models, with a context window of 16,000 tokens), enabling users to view or edit the equivalent of five average-sized Python files simultaneously, making it an ideal learning tool for beginners to take on larger challenges.