Fine-tuning

Overview

This screenshot showcases the interface for fine-tuning models within the Edge AI SDK. Fine-tuning is a technique used to optimize pre-trained models for specific tasks. By fine-tuning, you can improve the accuracy and performance of a model in a particular application using a smaller dataset.

Key Features

The interface provides three main fine-tuning options:

• Text-to-Text (Full Parameter): Finetune popular text-to-text models using full parameter adjustment. This method involves updating all parameters in the model, typically achieving the highest accuracy but requiring more computational resources and data. (Phison AI SSD required)

• Text-to-Text (LoRA): Finetune popular text-to-text models using LoRA (Low-Rank Adaptation). LoRA is a more efficient fine-tuning method that freezes the original weights of the pre-trained model and only trains a small number of additional parameters. This significantly reduces computational and memory requirements.

• Text-to-Image: Finetune popular text-to-image models. This option allows you to customize the output of text-to-image models for specific needs. This feature is under development and will be available in a future release.

Usage

Users can select the appropriate fine-tuning option based on their specific task and resources.

• For the highest possible accuracy when a Phison AI SSD is attached, use "Text-to-Text (Full Parameter).

• If computational resources are limited, or a more efficient fine-tuning process is desired, "Text-to-Text (LoRA)" can be used.

• If there is a need to customize the output of text-to-image models, "Text-to-Image" can be used.