Training Concepts#
This section provides conceptual material for training with Nemotron steps. Training Basics introduces fine-tuning approaches, tokenizers, the chat dataset format, and checkpoint layouts. For the step, configuration, and environment profile model that every Nemotron command shares, see Nemotron Steps Basics. The remaining pages explain how artifacts relate to one another and how the training libraries differ.
Defines supervised fine-tuning, parameter-efficient fine-tuning, reinforcement learning alignment, quantization, tokenizers, the chat dataset format, and checkpoints.
Explains how steps declare typed inputs and outputs and what common training and alignment paths look like.
Explains which library backs each step and how that choice determines data format, checkpoint layout, and parallelism.