Fine-Tuning Accuracy for Multiclass Classification - Ada model

Hello all,

I employed the ‘Ada’ model for a multiclass classification task with 40 unique labels. My training dataset consists of 40,000 incident descriptions and their corresponding incident labels. I performed fine-tuning 9 times on this training dataset. Each time, I achieved an accuracy of 76% when compared to the actual test dataset. Surprisingly, the accuracy remained consistent, without any noticeable increase or decrease. I’m perplexed about why I consistently obtain a 76% accuracy and would appreciate any insights or assistance in understanding this phenomenon. Thank you in advance for your help.