A public-health inspection desk has a mid-size instruction set and one workstation GPU. Full-weight supervised fine-tuning would update every parameter; LoRA would freeze the base and train low-rank adapters. Which experiment should they run first, and why?
Select an answer to reveal the explanation.
Short Explanation
Mid-size instruction set, one workstation GPU. Start with LoRA: it fits, and it keeps more of the base behavior intact than a full-weight rewrite. Full SFT is not required for every adaptation, and NIM is not a substitute for that choice.
Full Explanation
The choice between full supervised fine-tuning and LoRA is an experiment-design decision, not a product name. A mid-size instruction set on one GPU usually favors LoRA: fewer trainable weights, lower memory, and more of the base behavior left intact. Full-weight updates are not required for every adaptation, and serving is not a substitute for that choice. Associate practice starts with the cheaper, more conservative treatment unless the job truly needs a full rewrite.