twitter_RoBERTa_token_itr0_1e-05_essays_01_03_2022-14_58_58

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2698
• Precision: 0.3554
• Recall: 0.4884
• F1: 0.4114
• Accuracy: 0.8973

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	11	0.4423	0.0261	0.0184	0.0216	0.7728
No log	2.0	22	0.3220	0.1256	0.3129	0.1793	0.8735
No log	3.0	33	0.2561	0.2633	0.4264	0.3255	0.9103
No log	4.0	44	0.2535	0.3303	0.4509	0.3813	0.9115
No log	5.0	55	0.2414	0.3696	0.4693	0.4135	0.9181

Framework versions

• Transformers 4.15.0
• Pytorch 1.10.1+cu113
• Datasets 1.18.0
• Tokenizers 0.10.3