Customizations#

Customization, also known as fine-tuning, is the process of adapting a pre-trained base model to perform better on specific tasks or domains. Through NeMo Studio, you can configure and launch customization jobs, monitor their progress, and access the resulting custom models.

The Customizations page in NeMo Studio provides a centralized interface for managing your model fine-tuning jobs. You can create, monitor, and manage customization jobs that adapt base models to your specific use cases using your own datasets.

Backend Microservices#

In the backend, the UI communicates with NeMo Customizer to manage customization jobs and entities such as configurations, results, and metadata.

Customizations Page UI Overview#

The following are the main components and features of the Customizations page.

Customization Job Listing#

The Customizations page displays your customization jobs in a table format with the following columns:

  • Customization ID: The unique identifier for your customization job.

  • Description: Brief description or name of the customization.

  • Base Model: The foundation model being fine-tuned.

  • Dataset: The training dataset used for customization.

  • Finetuning: The fine-tuning method used.

  • Training: The training approach used.

  • Created: Timestamp showing when the job was created.

  • Time Elapsed: Duration of the completed job. Shows a hyphen (-) for jobs that have not completed.

  • Epochs: Shows the number of completed epochs out of total epochs with percentage.

  • Status: The current state of the job.

Search and Filter#

  • Search Jobs: Use the search bar to find customization jobs by name.

  • Filter Options: Apply filters to narrow down jobs based on status, base model, or other criteria.

  • Sort: Click column headers to sort jobs by name, status, or date.

Customization Job Management#

  • Create New Job: Start a new customization job by clicking the Customize Model button.

  • View Details: Access detailed information about each job, including configuration, logs, and metrics.

  • Job Actions: Use the context menu (three-dot menu icon) to perform actions such as canceling, restarting, or deleting jobs.

  • Monitor Progress: Track real-time progress of running customization jobs.

Customization Workflow#

The following are the common workflows for customizing models.

Create a Customization Job#

To create a new customization job:

  1. Navigate to the Customizations page from the left sidebar.

  2. Click the Create New Customization button.

  3. Select a base model to fine-tune.

  4. Choose a training dataset.

  5. Configure customization parameters such as learning rate, epochs, and batch size.

  6. Set a target for the customization job output.

  7. Review your configuration and submit the job.

Monitor Job Progress#

While a job is running, you can monitor its progress.

  • View real-time status updates of customization jobs on the Customizations page.

  • Click on a job to see the progress and details of each customization job.

Access Customized Models#

After a customization job completes successfully, you can access and use customized models for inference and evaluation.

  • Access the customized models in the Customizations page.

  • Use the customized models for prompt-tuning and inference in the Models page.

  • Use the customized models for evaluation in the Evaluations page.

Customization Methods#

NeMo Studio supports various customization methods backed by NeMo Customizer. To learn more about the customization methods, refer to Customization Concepts.