A school-cafeteria menu desk needs one base model that can tag allergen notes or draft tray labels without training a second full copy. How can a multi-task experiment do that?
Select an answer to reveal the explanation.
Short Explanation
One frozen base, two jobs: allergen tags and tray labels. Train a separate adapter per task and attach the matching one at eval time. A full fine-tune copy per task, a NIM endpoint per allergen word, and a vision-adapter detour miss that module design.
Full Explanation
Adapter-style PEFT supports multi-task experiments by swapping small modules on one frozen base. Each task keeps its own adapter instead of a second full fine-tune. Serving one endpoint per word, or moving the item to a multimodal exam, misses that module design. Associate practice stores the adapters and the shared base identity.