Arbi-Houssem/TunLangModel_test1.16

Whisper Tunisien

This model is a fine-tuned version of openai/whisper-small on the Tunisian_dataset_STT-TTS15s_filtred1.0_Mixed dataset. It achieves the following results on the evaluation set:

• Loss: 1.8540
• Wer: 90.7904

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-08
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.2076	4.5045	500	2.1992	120.4131
2.0416	9.0090	1000	2.0813	117.6732
1.9182	13.5135	1500	1.9770	119.2368
1.9273	18.0180	2000	1.9219	104.0023
1.8201	22.5225	2500	1.8901	94.8931
1.7708	27.0270	3000	1.8667	92.9709
1.7865	31.5315	3500	1.8565	90.8908
1.805	36.0360	4000	1.8540	90.7904

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1