Code-LMS

Code-LMS (featuring PolyCoder) is an open-source suite of large neural language models trained specifically on source code from 12 programming languages. It provides pretrained models of various sizes (160M, 0.4B, 2.7B parameters) for code generation and research, accessible via Huggingface, Docker images, or direct source. Code-LMS is aimed at researchers, engineers, and developers working on automated code generation, code understanding, and LLM evaluation in software contexts.