CodeGen2.5 от Salesforce доказывает, что в мире ИИ «меньше» часто значит «лучше»: эта компактная модель показывает SOTA-результаты в написании кода. Она идеально оптимизирована для интеграции в инструменты разработки, обеспечивая высокую скорость и точность генерации программного кода.
The family of Salesforce CodeGen models is growing with CodeGen2.5 – a small, but mighty model! While there has been a recent trend of large language models (LLM) of increasing size, we show that a small model can obtain surprisingly good performance, when being trained well. The key contributions towards productization of these models are: Releasing CodeGen2.5 LLM with state-of-the-art on HumanEval for 7B parameters. CodeGen2.5 with 7B is on par with >15B code-generation models (CodeGen1-16B, CodeGen2-16B, StarCoder-15B), less than half the size. Featuring robust infill sampling, that is, the model can “read” text of both the left and right hand size of the current position. Optimized for fast sampling under Flash attention for optimized serving and local deployment on personal machines. Permissively licensed in Apache 2.0.