Moroccan-Darija-STT-tiny-v1.6.2

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: 0.5041
• Wer: 265.2778
• Cer: 234.7702

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: 3.75e-05
• train_batch_size: 32
• eval_batch_size: 32
• 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: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 7

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.3599	0.3067	50	0.5879	342.3025	321.1932
1.0078	0.6135	100	0.5332	429.5515	391.1339
0.9529	0.9202	150	0.4855	384.4712	317.1326
0.8157	1.2270	200	0.4915	299.5984	252.5396
0.7735	1.5337	250	0.4780	385.3999	338.2396
0.8126	1.8405	300	0.4780	504.2169	452.5717
0.6253	2.1472	350	0.4674	237.0231	189.8620
0.6763	2.4540	400	0.4699	354.5683	302.8715
0.6164	2.7607	450	0.4630	219.2687	174.6702
0.5001	3.0675	500	0.4820	240.1523	194.6878
0.5454	3.3742	550	0.4760	206.9110	164.0120
0.4911	3.6810	600	0.4800	277.3009	233.3919
0.4934	3.9877	650	0.4891	237.7761	190.4988
0.5278	4.2945	700	0.4890	213.9224	178.0535
0.5464	4.6012	750	0.4933	251.3136	217.1512
0.4388	4.9080	800	0.4905	266.0894	224.1339
0.4061	5.2147	850	0.4947	227.1252	191.6372
0.4557	5.5215	900	0.4980	213.8638	173.6314
0.4852	5.8282	950	0.4970	253.1376	217.9467
0.393	6.1350	1000	0.5037	235.5589	198.1504
0.3682	6.4417	1050	0.5042	234.1449	199.3294
0.3934	6.7485	1100	0.5041	265.2778	234.7702

Framework versions

• Transformers 4.48.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0