Moroccan-Darija-STT-small-v1.6.13

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: 0.3369
• Wer: 79.3005
• Cer: 36.0370

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: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.5765	0.1775	30	0.4234	97.3645	51.8555
0.4537	0.3550	60	0.3846	94.5365	55.7354
0.4127	0.5325	90	0.3620	88.3283	42.6668
0.4127	0.7101	120	0.3442	93.3317	48.9604
0.4145	0.8876	150	0.3457	81.9946	38.7261
0.3463	1.0651	180	0.3407	93.4153	51.1461
0.3655	1.2426	210	0.3377	98.9207	57.9683
0.3476	1.4201	240	0.3346	85.2493	41.5773
0.3307	1.5976	270	0.3319	87.6757	44.5434
0.3538	1.7751	300	0.3248	84.4127	40.8577
0.3332	1.9527	330	0.3327	88.7801	48.9688
0.3298	2.1302	360	0.3355	84.3122	43.2596
0.3343	2.3077	390	0.3419	94.5448	51.6224
0.3238	2.4852	420	0.3344	89.3240	48.2948
0.2797	2.6627	450	0.3322	80.5723	37.7109
0.3209	2.8402	480	0.3324	79.0997	35.2415
0.3016	3.0178	510	0.3335	79.7272	36.6637
0.2911	3.1953	540	0.3356	75.3932	32.9814
0.2845	3.3728	570	0.3380	80.0368	37.5690
0.2791	3.5503	600	0.3359	77.4096	33.6216
0.2917	3.7278	630	0.3364	78.0873	34.4172
0.2752	3.9053	660	0.3369	79.3005	36.0370

Framework versions

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