abbenedekwhisper-base.en-finetuning3-D3K

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

• Loss: 3.3880
• Cer: 68.1692
• Wer: 115.5629
• Ser: 100.0
• Cer Clean: 3.6171
• Wer Clean: 6.2914
• Ser Clean: 7.0175

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-08
• train_batch_size: 16
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer	Ser	Cer Clean	Wer Clean	Ser Clean
7.3491	1.06	200	6.1358	64.7746	122.5166	100.0	3.2832	5.6291	7.0175
6.162	2.13	400	5.2935	64.2181	119.8675	100.0	3.7284	6.6225	7.8947
5.3192	3.19	600	4.7534	64.6633	119.2053	100.0	3.5058	6.2914	7.0175
4.7266	4.26	800	4.3761	65.1085	118.2119	100.0	3.2832	5.9603	6.1404
4.2728	5.32	1000	4.0472	65.9432	117.2185	100.0	3.2276	5.9603	6.1404
3.9248	6.38	1200	3.7904	66.7223	116.2252	100.0	3.2276	5.9603	6.1404
3.6714	7.45	1400	3.6008	67.8909	117.2185	100.0	3.1720	5.9603	6.1404
3.499	8.51	1600	3.4790	69.0595	118.2119	100.0	3.1720	5.9603	6.1404
3.393	9.57	1800	3.4106	68.9482	117.5497	100.0	3.1720	5.9603	6.1404
3.3491	10.64	2000	3.3880	68.1692	115.5629	100.0	3.6171	6.2914	7.0175

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu121
• Datasets 2.14.5
• Tokenizers 0.15.2