About Fine-Tuning

Learn how to fine-tune models by making requests to NVIDIA NeMo Customizer through the API. Fine-tuned models you have created can be deployed using NVIDIA NIMs.

Fine-Tuning Workflow

At a high level, the fine-tuning workflow consists of the following steps:

1. Create a Model Entity pointing to your base model checkpoint (stored as a FileSet).
2. Format a compatible dataset.
3. Create a customization job referencing the Model Entity.
4. Monitor the job until it completes.
5. The customization job automatically creates either:

• LoRA jobs: An adapter attached to the original Model Entity
• Full fine-tuning jobs: A new Model Entity with the customized weights

1. Deploy the model using the Deployment Management Service.
2. Move on to Evaluate the output model.

Container Images

Fine-tuning jobs run in container images published to NVIDIA NGC. Use the image tag that matches your NeMo Helix release:

These public images can be pulled directly from nvcr.io:

docker pull nvcr.io/nvidia/nemo-helix/nhx-customizer-tasks:<image-tag>
docker pull nvcr.io/nvidia/nemo-helix/nhx-automodel-training:<image-tag>
docker pull nvcr.io/nvidia/nemo-helix/nhx-unsloth-training:<image-tag>

Most users do not need to pull these images manually. NeMo Helix resolves them from the configured platform image registry and tag. Pull or mirror them when you are operating a self-managed deployment, preloading an air-gapped environment, or validating registry access. If your environment requires authenticated registry pulls, authenticate to nvcr.io with an NGC API key before pulling or mirroring the images.

Model Catalog

Explore the model families and sizes supported by NVIDIA NeMo Customizer.

Task Guides

Perform common fine-tuning tasks.

Tutorials

Follow these tutorials to learn how to accomplish common fine-tuning tasks.

References