disaster-tweet-5

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base-sentiment on a Kaggle competition dataset. It achieves the following results on the evaluation set:

Loss: 0.4107

Model description

More information needed

Intended uses & limitations

This model was created for the Natural Language Processing with Disaster Tweets Kaggle competition.

Training and evaluation data

Information about the data can be found here

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 3e-5
train_batch_size: 64
eval_batch_size: 64
seed: 7
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
0.7065	0.12	12	0.7027
0.7021	0.25	24	0.6997
0.7048	0.38	36	0.6948
0.7025	0.5	48	0.6887
0.683	0.62	60	0.6808
0.6748	0.75	72	0.6715
0.6701	0.88	84	0.6610
0.6564	1.0	96	0.6487
0.6418	1.12	108	0.6341
0.6268	1.25	120	0.6165
0.6362	1.38	132	0.5985
0.5824	1.5	144	0.5776
0.5766	1.62	156	0.5541
0.5417	1.75	168	0.5281
0.5232	1.88	180	0.5064
0.4737	2.0	192	0.4909
0.4479	2.12	204	0.4826
0.456	2.25	216	0.4662
0.4718	2.38	228	0.4541
0.4198	2.5	240	0.4451
0.4333	2.62	252	0.4376
0.4086	2.75	264	0.4337
0.4419	2.88	276	0.4332
0.3857	3.0	288	0.4225
0.3878	3.12	300	0.4188
0.3578	3.25	312	0.4280
0.3562	3.38	324	0.4234
0.4125	3.5	336	0.4147
0.3882	3.62	348	0.4090
0.3751	3.75	360	0.4145
0.3892	3.88	372	0.4077
0.3946	4.0	384	0.4107

Framework versions

Transformers 4.28.0
Pytorch 1.13.0
Datasets 2.1.0
Tokenizers 0.13.2