MindPet (Pet: Parameter-Efficient Tuning) is a suite of fine-tuning algorithms belonging to the Mindspore domain. With the continuous increase of computing power, the unlimited potential of large models has also been tapped. However, it has brought huge costs in application and training, making commercial implementation difficult. Therefore, a new parameter-efficient algorithm emerges. Compared with the standard full-parameter fine-tuning, these algorithms only need to fine-tune a small part of the parameters, which can greatly reduce the calculation and storage costs, while comparable to the performance of the full-parameter fine-tuning. . Fine-tuning algorithm API Currently MindPet has provided the following five classic low-parameter… |
#Finetuning #algorithm #suite #MindPet