Bert-Sentiment-Fa

This model is a fine-tuned version of HooshvareLab/bert-fa-base-uncased on the Neshan Sentiment dataset. It achieves the following results on the evaluation set:

• Loss: 0.4558
• Accuracy: 0.8549
• F1: 0.7877

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-06
• train_batch_size: 16
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
No log	1.0	143	0.6143	0.8078	0.6887
No log	2.0	286	0.4310	0.8392	0.7556
No log	3.0	429	0.4002	0.8627	0.8037
0.5214	4.0	572	0.3933	0.8588	0.7984
0.5214	5.0	715	0.4046	0.8588	0.7945
0.5214	6.0	858	0.4169	0.8588	0.7990
0.2054	7.0	1001	0.4345	0.8549	0.7848
0.2054	8.0	1144	0.4277	0.8627	0.8001
0.2054	9.0	1287	0.4509	0.8588	0.7918
0.2054	10.0	1430	0.4558	0.8549	0.7877

Framework versions

• Transformers 4.33.0
• Pytorch 2.0.0
• Datasets 2.1.0
• Tokenizers 0.13.3