NeMo Launcher (deprecated)

This legacy integration is deprecated. Use Nemo Run for NeMo training and fine-tuning, or MegatronBridge for a maintained training starter.

Migrate an experiment

Choose the matching model and training recipe in the replacement workload. Transfer dataset and tokenizer paths, parallelism, and batch settings deliberately; the launcher’s configuration is not interchangeable with the new recipe. Run a short case and compare its output before scaling up.

Dataset preparation

Provision the selected recipe’s data on storage accessible to the execution nodes. Use that recipe’s documented format and tokenizer, then configure paths and mounts. See Prepare model weights and Set up your execution environment.