distilroberta-topic-classification

This model is a fine-tuned version of distilroberta-topic-base on a dataset of headlines. It achieves the following results on the evaluation set:

• Loss: 2.235735
• F1: 0.756

Training and evaluation data

The following data sources were used:

• 22k News articles classified into 120 different topics from Hugging face

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 12345
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 16
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1
2.3851	1.0	561	2.3445	0.6495
2.1441	2.0	1122	2.1980	0.7019
1.9992	3.0	1683	2.1720	0.7189
1.8384	4.0	2244	2.1425	0.7403
1.7468	5.0	2805	2.1666	0.7453
1.6360	6.0	3366	2.1779	0.7456
1.5935	7.0	3927	2.2003	0.7555
1.5460	8.0	4488	2.2157	0.7575
1.5510	9.0	5049	2.2300	0.7536
1.5097	10.0	5610	2.2357	0.7547

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0
• Datasets 2.15.0
• Tokenizers 0.15.0