ES-ENG-mDeBERTa-sentiment

This model is a fine-tuned version of microsoft/mdeberta-v3-base on a Custom dataset.

The best model (stopped after 20 epochs) achieves the following results on the evaluation set:

• Loss:0.7549
• Accuracy: 0.6806
• F1: 0.6783
• Precision: 0.6775
• Recall: 0.6806

Intended uses & limitations

Note that commercial use with this model is prohibited.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-06
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
1.0833	1.0	208	1.0218	0.3835	0.2413	0.4619	0.3835
0.9826	2.0	416	0.9220	0.5673	0.5524	0.5583	0.5673
0.8951	3.0	624	0.8435	0.5916	0.5836	0.5853	0.5916
0.8377	4.0	832	0.8200	0.5994	0.5916	0.5931	0.5994
0.8083	5.0	1040	0.8110	0.5977	0.5960	0.5957	0.5977
0.7854	6.0	1248	0.8009	0.5994	0.5943	0.5916	0.5994
0.7699	7.0	1456	0.7919	0.6084	0.6065	0.6055	0.6084
0.7489	8.0	1664	0.7827	0.6230	0.6164	0.6192	0.6230
0.7323	9.0	1872	0.7739	0.6272	0.6251	0.6246	0.6272
0.7162	10.0	2080	0.7725	0.6408	0.6332	0.6351	0.6408
0.6958	11.0	2288	0.7636	0.6414	0.6393	0.6383	0.6414
0.6816	12.0	2496	0.7582	0.6495	0.6491	0.6516	0.6495
0.6706	13.0	2704	0.7492	0.6628	0.6607	0.6595	0.6628
0.6569	14.0	2912	0.7554	0.6586	0.6580	0.6579	0.6586
0.6422	15.0	3120	0.7525	0.6676	0.6661	0.6667	0.6676
0.6408	16.0	3328	0.7527	0.6660	0.6658	0.6669	0.6660
0.6273	17.0	3536	0.7483	0.6712	0.6700	0.6710	0.6712
0.6186	18.0	3744	0.7531	0.6748	0.6731	0.6727	0.6748
0.6107	19.0	3952	0.7482	0.6799	0.6786	0.6782	0.6799
0.6055	20.0	4160	0.7549	0.6806	0.6783	0.6775	0.6806
0.6026	21.0	4368	0.7574	0.6725	0.6719	0.6733	0.6725
0.5906	22.0	4576	0.7587	0.6728	0.6721	0.6723	0.6728
0.5888	23.0	4784	0.7621	0.6761	0.6756	0.6758	0.6761

Framework versions

• Transformers 4.29.2
• Pytorch 2.0.1+cu117
• Datasets 2.12.0
• Tokenizers 0.13.3