Hi, I am using tf-xlm-r-base model for a sentiment classification (multi-class) task with 4 classes. I used both trainer() api and keras native method. Initially, I got some acceptable result but later it predicts only one class for the same data set. I am following this guide . Below are my outputs…

XLM-R classifier predictions produce errors

vinurad13 June 25, 2021, 7:06am 2

also found that for binary classification, (two from above four classes), it behaves similarly. And tried changing labels to float type and different loss function too.

Topic		Replies	Views
Xlm-roberta-base predicting always same class, other models don't Intermediate	2	1110	June 7, 2023
XLM-Roberta for many-topic classification Beginners	1	1176	December 31, 2021
Model predicts only one label on a multi-label Text Classification task (XLMRoberta) Beginners	0	455	August 14, 2023
Fine-tuning XLM-RoBERTa for binary sentiment classification Beginners	1	1441	November 4, 2021
Fine-Tune Xlm-roberta-large-xnli 🤗Transformers	1	1935	December 28, 2021