Moroccan-Darija-STT-large-turbo-v1.6.2

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3670
• Wer: 93.8922
• Cer: 53.0699

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: 4.375e-06
• 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
0.9764	0.3067	50	0.4338	95.5238	48.4807
0.7447	0.6135	100	0.3480	76.8742	31.1994
0.7076	0.9202	150	0.3278	82.1703	38.6822
0.6132	1.2270	200	0.3237	77.6606	34.7094
0.562	1.5337	250	0.3166	73.6780	31.1944
0.6057	1.8405	300	0.3114	91.5997	47.9874
0.4726	2.1472	350	0.3175	81.8859	38.8882
0.5005	2.4540	400	0.3188	85.1657	41.4337
0.4421	2.7607	450	0.3149	82.8397	40.7699
0.3551	3.0675	500	0.3303	105.1456	64.8954
0.3845	3.3742	550	0.3288	81.1747	38.4896
0.3513	3.6810	600	0.3292	83.2580	41.5519
0.3658	3.9877	650	0.3317	99.4896	54.9820
0.4258	4.2945	700	0.3384	89.1064	46.7882
0.4206	4.6012	750	0.3433	98.1844	56.1390
0.3115	4.9080	800	0.3406	89.1315	48.7577
0.3037	5.2147	850	0.3473	76.4809	38.0487
0.35	5.5215	900	0.3525	85.2410	44.4437
0.3754	5.8282	950	0.3535	97.1888	58.1035
0.2915	6.1350	1000	0.3661	94.5616	54.5327
0.2725	6.4417	1050	0.3676	99.8494	58.5798
0.2822	6.7485	1100	0.3670	93.8922	53.0699

Framework versions

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