Moroccan-Darija-STT-small-v1.6.11

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

• Loss: 0.6266
• Wer: 98.5442
• Cer: 85.3893

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: 1.25e-05
• train_batch_size: 128
• eval_batch_size: 128
• 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_steps: 10
• num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3624	0.3030	30	0.6341	100.3430	92.3044
0.3091	0.6061	60	0.5572	98.1760	83.6191
0.2958	0.9091	90	0.5413	98.3350	84.3521
0.2854	1.2121	120	0.5639	98.9207	88.2523
0.276	1.5152	150	0.6350	98.8119	88.2742
0.2764	1.8182	180	0.6088	98.7199	86.9500
0.2609	2.1212	210	0.6280	98.5776	86.2541
0.2658	2.4242	240	0.6107	98.6529	86.9973
0.2494	2.7273	270	0.6355	99.0629	88.0226
0.2419	3.0303	300	0.6267	98.9709	86.8537
0.2458	3.3333	330	0.6331	99.0295	85.8605
0.2273	3.6364	360	0.6155	98.3517	83.8893
0.2491	3.9394	390	0.6208	98.8788	86.0531
0.2419	4.2424	420	0.6260	98.6780	82.9046
0.2387	4.5455	450	0.6294	98.8454	85.9652
0.2282	4.8485	480	0.6209	98.7199	86.4129
0.228	5.1515	510	0.6260	99.0880	85.6730
0.2403	5.4545	540	0.6504	99.1801	85.9720
0.227	5.7576	570	0.6266	98.5442	85.3893

Framework versions

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