MubarakB/results

Helsinki_lg_inf_en

This model is a fine-tuned version of MT-Informal-Languages/Helsinki-NLP-opus-mt-ug on the Luganda Formal Data dataset. It achieves the following results on the evaluation set:

• Loss: 0.0505
• Bleu: 57.3885
• Gen Len: 17.3595

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
No log	1.0	153	0.4708	0.9369	19.2244
No log	2.0	306	0.4227	1.1005	20.8706
No log	3.0	459	0.3854	1.4207	19.8702
1.1563	4.0	612	0.3519	1.7877	19.5442
1.1563	5.0	765	0.3216	2.4366	18.7977
1.1563	6.0	918	0.2929	3.0827	18.6413
0.375	7.0	1071	0.2677	3.9367	19.2035
0.375	8.0	1224	0.2427	5.605	18.5111
0.375	9.0	1377	0.2192	7.0359	18.6204
0.2959	10.0	1530	0.1980	9.5819	17.8284
0.2959	11.0	1683	0.1794	11.9364	17.7428
0.2959	12.0	1836	0.1621	13.9353	18.0643
0.2959	13.0	1989	0.1464	16.9189	17.8714
0.2334	14.0	2142	0.1315	19.2848	18.0201
0.2334	15.0	2295	0.1189	22.6041	17.973
0.2334	16.0	2448	0.1085	25.554	18.0324
0.1848	17.0	2601	0.0992	28.6049	17.4644
0.1848	18.0	2754	0.0905	31.9759	17.8104
0.1848	19.0	2907	0.0828	35.5846	17.8108
0.1507	20.0	3060	0.0764	39.748	17.656
0.1507	21.0	3213	0.0712	42.3511	17.5602
0.1507	22.0	3366	0.0665	45.7843	17.5238
0.1285	23.0	3519	0.0628	48.4047	17.5233
0.1285	24.0	3672	0.0592	50.5559	17.3403
0.1285	25.0	3825	0.0564	52.0378	17.4443
0.1285	26.0	3978	0.0545	54.0726	17.579
0.1132	27.0	4131	0.0526	55.201	17.4017
0.1132	28.0	4284	0.0515	56.6048	17.4447
0.1132	29.0	4437	0.0508	57.2182	17.448
0.1047	30.0	4590	0.0505	57.3885	17.3595

Framework versions

• Transformers 4.42.3
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1