NeMo

NVIDIA NeMo is a GPU-accelerated Python toolkit for building, training, and fine-tuning state-of-the-art generative AI models across speech, NLP, and multimodal domains.

Engineers use NeMo to develop production-ready conversational AI through a modular framework built on PyTorch and Lightning. It provides pre-trained checkpoints and collections (ASR, NLP, and TTS) that scale across multi-node clusters using Megatron-LM for massive Large Language Models. By integrating NVIDIA's TensorRT-LLM and Triton Inference Server, the toolkit streamlines the pipeline from initial data curation to high-throughput deployment.

https://github.com/NVIDIA/NeMo