nemo-customizer

You can automate SFT and LoRA fine-tuning by orchestrating dataset preparation, model registration, and GPU-accelerated Docker job submission through the NeMo CLI. This removes manual infrastructure management and streamlines the entire training lifecycle.

Yes, NeMo customizer supports both Automodel and Unsloth backends for efficient SFT and LoRA fine-tuning. It handles the integration natively, allowing you to submit training jobs directly through the platform's execution profiles.

The best way to manage datasets and model registration is using the NeMo CLI for dataset conversion, fileset management, and model entity registration. This ensures your custom datasets are correctly prepared before submitting GPU-accelerated training jobs.

Yes, NeMo customizer handles job polling to monitor training progress and manages the lifecycle through to adapter registration. It ensures reliable training and seamless deployment of LoRA adapters directly on the NeMo Platform.

Yes, you need platform-specific execution profiles to ensure reliable GPU-accelerated Docker job submission for SFT and LoRA. These profiles integrate with the NeMo CLI to manage the underlying training infrastructure automatically.