Supervised Fine-Tuning Steps

This section documents the supervised fine-tuning (SFT) steps registered under src/nemotron/steps/sft/. The two steps target different training libraries and consume different data formats. Both produce checkpoints you can use as warm-start policies for reinforcement learning alignment.

Steps

sft/automodel

Supervised fine-tuning with the NeMo AutoModel library against Hugging Face base models and JSON Lines chat datasets.

reference

sft/automodel

sft/megatron_bridge

Supervised fine-tuning on top of NVIDIA Megatron-Bridge for distributed training of the Nemotron model family.

reference

sft/megatron_bridge

Related Documentation

• Nemotron Steps CLI Reference covers the shared option set, dotlist overrides, and passthrough arguments.

• Choose an SFT Backend compares sft/automodel and sft/megatron_bridge.

• Configuration Conventions describes the per-step config/ layout.