Teach AI to Code in Every Language with NVIDIA NeMo | NVIDIA GTC

Large language models for code are transforming how developers create, maintain, and understand software—but building these models from the ground up requires the right tools and know-how. In this talk, we’ll explore how to leverage the NVIDIA NeMo framework and NVIDIA’s accelerated computing infrastructure to train state-of-the-art (SOTA) coding LLMs capable of handling multiple programming languages. Gain practical insights into dataset preparation, multilingual training challenges, and effective strategies for balancing diverse code sources. We’ll cover workflow design for large-scale distributed training, discussing optimization techniques and best practices for scaling across GPU clusters. The session will also highlight real-world applications of multilingual coding LLMs—from powering intelligent developer assistants to managing complex enterprise codebases. Miguel Martinez | Sr. Applied Deep Learning Researcher | NVIDIA Meriem Bendris | Sr. Deep Learning Data Scientist | NVIDIA Oleg Sudakov | Solutions Architect | NVIDIA Key Takeaways: A practical understanding of how to use NVIDIA NeMo and NVIDIA infrastructure to train SOTA coding LLMs Insights into dataset preparation, multilingual training challenges, and strategies to overcome them Knowledge of optimization techniques and best practices for scaling training across large compute environments Awareness of real-world applications for multilingual coding LLMs, from developer assistance to enterprise codebase management A clear roadmap for experimenting with and deploying advanced coding LLMs using NVIDIA’s technology stack Industry: All Industries Topic: Agentic AI / Generative AI - Code / Software Generation Technical Level: Technical - Intermediate Intended Audience: Data Scientist NVIDIA Technology: DGX Platform, TensorRT, NeMo, Triton, NVIDIA NIM, NVIDIA AI Enterprise, DGX Cloud, Nemotron #nvidiagtc