NeMo Megatron is an end-to-end containerized framework that allows developers to efficiently train and deploy language models with billions and trillions of parameters. NeMo Megatron delivers high training efficiency across thousands of GPUs and makes it practical for enterprises to build and deploy large-scale models. It provides capabilities to curate training data, train large-scale models up to trillions of parameters, customize using prompt learning techniques and deploy using NVIDIA Triton™ Inference Server to run large-scale models on multiple GPUs and multiple nodes. This deep learning software stack is optimized for NVIDIA DGX SuperPODs™ using NVIDIA InfiniBand, Microsoft Azure, Oracle Cloud Infrastructure, and Amazon Web Services for training and inferring complex LLM workloads.
After a successful early-access program, we are now making NeMo Megatron available through open beta to everyone who wants to train and deploy GPT-3, T5, and mT5 style models to perform tasks such as answering deep domain questions, text completion, comprehending, and summarizing complex documents, etc.
To use, please fill out the short form and provide details about your use case. You must be a member of the NVIDIA Developer Program and logged in with your organization's email address.