Developer Webinar
Building and Deploying Multi-Language AI Models With DeepL and NVIDIA
With breakthrough advances in hardware, model architectures, and enterprise-ready solutions, translation and generative language technology are advancing with unprecedented speed, accuracy, and flexibility.
Discover how DeepL deployed NVIDIA’s cutting-edge DGX SuperPOD to accelerate training and deployment of LLMs for translation, which includes real-time speech and direct translation without needing to translate to English. Hear the latest training and inference tricks, such as FP8 and FP4 quantization, LoRA adapters, and mixture of experts (MoE), along with DeepL’s application of NVIDIA TensorRT-LLM to deliver blazing-fast, scalable performance.
In this session, you’ll get practical knowledge of how improvements in one language enhance others, how custom rules empower business use cases, and why scaling to massive models now means fewer infrastructural headaches and greater efficiency in global language solutions.
Learnings
In this webinar you’ll learn how to:
- How DeepL achieved more accurate, real-time AI translation and advanced voice capabilities through their collaboration with NVIDIA?
- Leverage innovative training approaches such as FP8 precision, LoRA adapters, and mixture of experts (MoE) to boost global translation quality and efficiently transfer improvements across languages.
- Use NVIDIA TensorRT-LLM and quantization methods to accelerate translation inference, enabling large models to run faster and reduce infrastructure complexity.
- Apply advanced post-editing or customization features to resolve translation ambiguity, enforce preferred terms, and better serve enterprise and business needs.
- Capitalize on scaling breakthroughs from NVIDIA DGX SuperPOD, empowering DeepL to translate massive data sets—such as the entire internet—in days, making new large-scale applications possible.
- Gain insights into current and future roadmaps for model training and inference to understand both the business value and technical innovation underlying next-generation translation AI.
