ethiopic-sec2sec-tigrinya

This model is a fine-tuned version of Samuael/ethiopic-sec2sec-tigrinya on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 4.3390
• Wer: 0.2570
• Cer: 0.1269
• Bleu: 67.2562

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: 0.0002
• train_batch_size: 64
• eval_batch_size: 128
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 9
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Bleu
0.086	1.0	433	4.0475	0.2592	0.1222	67.0097
0.1177	2.0	866	3.8429	0.3645	0.2577	59.7600
0.1312	3.0	1299	4.0294	0.2630	0.1337	66.8126
0.1322	4.0	1732	4.1292	0.2527	0.1231	68.4884
0.1411	5.0	2165	3.9989	0.2662	0.1358	66.4299
0.1686	6.0	2598	4.1555	0.2592	0.1388	67.9808
0.1587	7.0	3031	4.2802	0.2543	0.1264	66.9025
0.1386	8.0	3464	4.2682	0.2554	0.1283	68.7450
0.0886	9.0	3897	4.3390	0.2570	0.1269	67.2562

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3