afri-byt5-base

This model is a fine-tuned version of masakhane/afri-byt5-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6309
• Bleu: 0.8581
• Rouge: 0.209
• Meteor: 0.1002
• Chrf++: 11.7747

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: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Rouge	Meteor	Chrf++
1.7895	1.0331	500	1.4756	0.1421	0.1368	0.0423	6.6274
1.231	2.0661	1000	1.1196	0.3613	0.1524	0.0566	7.8681
1.0757	3.0992	1500	1.0213	0.4626	0.1613	0.0643	8.7803
0.9816	4.1322	2000	0.9547	0.5325	0.1663	0.0683	9.2492
0.9185	5.1653	2500	0.9079	0.583	0.1705	0.0723	9.629
0.8689	6.1983	3000	0.8699	0.6365	0.1754	0.0764	9.9206
0.8289	7.2314	3500	0.8400	0.6422	0.1767	0.0773	10.1022
0.7979	8.2645	4000	0.8188	0.6831	0.1798	0.0792	10.2409
0.7708	9.2975	4500	0.7943	0.7081	0.1831	0.0819	10.4372
0.7484	10.3306	5000	0.7783	0.724	0.185	0.0837	10.5772
0.7269	11.3636	5500	0.7613	0.7353	0.1865	0.0845	10.6582
0.7079	12.3967	6000	0.7459	0.7397	0.1888	0.0865	10.8046
0.6935	13.4298	6500	0.7388	0.7697	0.1907	0.0882	10.875
0.6792	14.4628	7000	0.7236	0.753	0.1917	0.0887	10.9335
0.6657	15.4959	7500	0.7149	0.7707	0.1922	0.0891	10.9879
0.6547	16.5289	8000	0.7055	0.7651	0.1925	0.0891	11.0083
0.6422	17.5620	8500	0.7007	0.7991	0.1953	0.0916	11.13
0.6321	18.5950	9000	0.6907	0.8011	0.1961	0.0919	11.1865
0.6219	19.6281	9500	0.6840	0.7862	0.1962	0.0917	11.1917
0.6152	20.6612	10000	0.6802	0.7879	0.1967	0.0925	11.2574
0.607	21.6942	10500	0.6731	0.7999	0.1973	0.0934	11.2744
0.5988	22.7273	11000	0.6683	0.8043	0.1976	0.0935	11.3227
0.5925	23.7603	11500	0.6645	0.8044	0.1996	0.0946	11.3835
0.5867	24.7934	12000	0.6590	0.8022	0.2005	0.095	11.3975
0.5799	25.8264	12500	0.6575	0.8159	0.2008	0.0951	11.4052
0.5734	26.8595	13000	0.6524	0.8192	0.2011	0.0956	11.4084
0.5699	27.8926	13500	0.6493	0.8098	0.2021	0.096	11.4505
0.5643	28.9256	14000	0.6444	0.8235	0.2023	0.0964	11.4778
0.56	29.9587	14500	0.6431	0.8246	0.2027	0.0968	11.5121
0.5547	30.9917	15000	0.6414	0.8263	0.2037	0.0967	11.5245
0.5517	32.0248	15500	0.6388	0.8318	0.2039	0.0976	11.5743
0.5475	33.0579	16000	0.6368	0.8329	0.2044	0.098	11.5694
0.5442	34.0909	16500	0.6359	0.8268	0.2054	0.0985	11.5693
0.5413	35.1240	17000	0.6312	0.8351	0.2049	0.0982	11.6073
0.5387	36.1570	17500	0.6315	0.8304	0.2063	0.0988	11.5939
0.5357	37.1901	18000	0.6311	0.834	0.2064	0.0991	11.633
0.533	38.2231	18500	0.6279	0.8371	0.2066	0.0991	11.6246
0.5299	39.2562	19000	0.6276	0.8357	0.207	0.0989	11.6292
0.5302	40.2893	19500	0.6243	0.8376	0.2071	0.0993	11.6437
0.527	41.3223	20000	0.6236	0.8406	0.2068	0.0992	11.6423
0.5248	42.3554	20500	0.6237	0.839	0.2075	0.0993	11.6627
0.5221	43.3884	21000	0.6239	0.8354	0.2068	0.0993	11.6636
0.5239	44.4215	21500	0.6232	0.8339	0.2076	0.0999	11.6581
0.5207	45.4545	22000	0.6215	0.835	0.2081	0.0999	11.6802
0.5191	46.4876	22500	0.6214	0.8355	0.2073	0.0997	11.6884
0.519	47.5207	23000	0.6223	0.8362	0.2075	0.0997	11.6955
0.5188	48.5537	23500	0.6216	0.8375	0.2071	0.0998	11.698
0.5195	49.5868	24000	0.6215	0.835	0.2075	0.0998	11.692

Framework versions

• Transformers 4.57.6
• Pytorch 2.11.0+cu130
• Datasets 4.8.5
• Tokenizers 0.22.2