Granite Code Models: A Family of Open Foundation Models for Code Intelligence

• DOI: 10.48550/arxiv.2405.04324
• Publication Date: 2024-05-07
• Authors: Mayank Mishra; Matt Stallone; Gaoyuan Zhang; Yikang Shen; Aditya Prasad; Adriana Meza Soria; Michele Merler; Parameswaran Selvam; Saptha Surendran; Shivdeep Singh; Manish Sethi; Xuan-Hong Dang; Pengyuan Li; Kun‐Lung Wu; Syed Zawad; Andrew Coleman; Matthew White; Mark Lewis; Raju Pavuluri; Yan Koyfman; Boris Lublinsky; Maximilien de Bayser; Ibrahim Abdelaziz; Kinjal Basu; Mayank Agarwal; Yi Zhou; Chris Johnson; Aanchal Goyal; Hima Patel; Yousaf Shah; Petros Zerfos; Heiko Ludwig; Asim Munawar; Maxwell Crouse; Pavan Kapanipathi; Shweta Salaria; Bob Calio; Sophia Wen; Seetharami Seelam; Brian Belgodere; Carlos Roberto Fo...; Amith Singhee; Nirmit Desai; David Cox; Ruchir Puri; Rameswar Panda
• Keywords: Code (set theory); Foundation (evidence)

Large Language Models (LLMs) trained on code are revolutionizing the software development process. Increasingly, LLMs being integrated into environments to improve productivity of human programmers, and LLM-based agents beginning show promise for handling complex tasks autonomously. Realizing full potential requires a wide range capabilities, including generation, fixing bugs, explaining documenting code, maintaining repositories, more. In this work, we introduce Granite series decoder-only models generative tasks, with written in 116 programming languages. The Code family consists ranging size from 3 34 billion parameters, suitable applications application modernization on-device memory-constrained use cases. Evaluation comprehensive set demonstrates that consistently reaches state-of-the-art performance among available open-source LLMs. model was optimized enterprise workflows performs well across coding (e.g. explanation), making it versatile all around model. We release our under an Apache 2.0 license both research commercial use.