DistilBERTFINAL_ctxSentence_TRAIN_webDiscourse_TEST_NULL_second_train_set_null_False

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: 3.2555
• Precision: 1.0
• Recall: 0.0200
• F1: 0.0393
• Accuracy: 0.0486

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	95	0.5756	nan	0.0	nan	0.715
No log	2.0	190	0.5340	0.6429	0.1579	0.2535	0.735
No log	3.0	285	0.5298	0.5833	0.3684	0.4516	0.745
No log	4.0	380	0.5325	0.5789	0.3860	0.4632	0.745
No log	5.0	475	0.5452	0.4815	0.4561	0.4685	0.705

Framework versions

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