whisper-small-CV_Fleurs_AMMI_ALFFA-sw-50hrs-v1

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5447
• Wer: 0.2171
• Cer: 0.0854

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_hf 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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.2946	1.0	1925	0.5694	0.4428	0.1833
0.7795	2.0	3850	0.4233	0.2721	0.0984
0.4733	3.0	5775	0.3853	0.2346	0.0863
0.3061	4.0	7700	0.3903	0.2398	0.0909
0.221	5.0	9625	0.4062	0.2204	0.0799
0.1851	6.0	11550	0.4316	0.2096	0.0750
0.1674	7.0	13475	0.4491	0.3117	0.1553
0.1615	8.0	15400	0.4725	0.2198	0.0826
0.1622	9.0	17325	0.4878	0.2335	0.0905
0.1609	10.0	19250	0.5068	0.2398	0.0983
0.1487	11.0	21175	0.5208	0.2377	0.0935
0.1294	12.0	23100	0.5156	0.2429	0.1048
0.1102	13.0	25025	0.5297	0.2193	0.0858
0.0944	14.0	26950	0.5277	0.2175	0.0831
0.0843	15.0	28875	0.5454	0.2095	0.0798
0.0761	16.0	30800	0.5447	0.2171	0.0854

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1