xlm-roberta-large-azsci-topics

This model is a fine-tuned version of FacebookAI/xlm-roberta-large on azsci_topics dataset. It achieves the following results on the evaluation set:

• Loss: 0.4012
• Precision: 0.9115
• Recall: 0.9158
• F1: 0.9121
• Accuracy: 0.9158

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• 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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	288	0.6402	0.8063	0.8073	0.7900	0.8073
1.0792	2.0	576	0.4482	0.8827	0.8776	0.8743	0.8776
1.0792	3.0	864	0.3947	0.8968	0.9019	0.8977	0.9019
0.3135	4.0	1152	0.4177	0.9043	0.9080	0.9047	0.9080
0.3135	5.0	1440	0.4012	0.9115	0.9158	0.9121	0.9158

Evaluation results

Topic	Precision	Recall	F1	Support
Aqrar elmlər	0.846154	0.814815	0.830189	27
Astronomiya	0.666667	1	0.8	2
Biologiya elmləri	0.910891	0.87619	0.893204	105
Coğrafiya	0.888889	0.941176	0.914286	17
Filologiya elmləri	0.971098	0.96	0.965517	175
Fizika	0.769231	0.882353	0.821918	34
Fəlsəfə	0.875	0.5	0.636364	14
Hüquq elmləri	0.966667	1	0.983051	29
Kimya	0.855072	0.967213	0.907692	61
Memarlıq	0.714286	1	0.833333	5
Mexanika	0	0	0	4
Pedaqogika	0.958333	0.978723	0.968421	47
Psixologiya	0.944444	0.944444	0.944444	18
Riyaziyyat	0.921053	0.897436	0.909091	39
Siyasi elmlər	0.785714	0.88	0.830189	25
Sosiologiya	0.666667	1	0.8	4
Sənətşünaslıq	0.84	0.893617	0.865979	47
Tarix	0.933333	0.897436	0.915033	78
Texnika elmləri	0.894737	0.817308	0.854271	104
Tibb elmləri	0.935484	0.97973	0.957096	148
Yer elmləri	0.846154	0.846154	0.846154	13
İqtisad elmləri	0.973684	0.973684	0.973684	152
Əczaçılıq elmləri	0	0	0	4
macro avg	0.78972	0.828273	0.80217	1152
weighted avg	0.911546	0.915799	0.912067	1152

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2