b1n1yam/el-tiny

tiny Amharic - Biniyam Daniel

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0803
• Wer: 24.4366

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5795	0.0665	100	1.5182	150.6407
1.2873	0.1330	200	1.2244	109.2797
0.3664	0.1995	300	0.3282	70.2607
0.2133	0.2660	400	0.2056	52.2095
0.171	0.3324	500	0.1731	46.8184
0.1422	0.3989	600	0.1467	41.7587
0.1392	0.4654	700	0.1344	38.8643
0.1264	0.5319	800	0.1247	36.7654
0.1113	0.5984	900	0.1186	34.3570
0.1038	0.6649	1000	0.1125	33.0977
0.0978	0.7314	1100	0.1091	33.0535
0.0959	0.7979	1200	0.1033	30.3137
0.0876	0.8644	1300	0.1003	29.6730
0.0975	0.9309	1400	0.0966	29.6730
0.0863	0.9973	1500	0.0968	28.9660
0.0633	1.0638	1600	0.0934	28.0601
0.0612	1.1303	1700	0.0913	28.3031
0.0576	1.1968	1800	0.0905	27.1542
0.0652	1.2633	1900	0.0886	27.1984
0.0636	1.3298	2000	0.0857	26.5135
0.0623	1.3963	2100	0.0852	25.9611
0.0556	1.4628	2200	0.0839	25.3646
0.0569	1.5293	2300	0.0827	25.6739
0.0574	1.5957	2400	0.0822	25.2762
0.0649	1.6622	2500	0.0813	24.9448
0.0744	1.7287	2600	0.0808	24.7238
0.0686	1.7952	2700	0.0805	24.7017
0.0587	1.8617	2800	0.0803	24.4145
0.0615	1.9282	2900	0.0803	24.4366
0.0637	1.9947	3000	0.0803	24.4366

Framework versions

• Transformers 4.57.1
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.22.1