whisper-small-en

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: 0.0947
• Wer: 11.3030

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-05
• train_batch_size: 16
• 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: 50
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0	7.6923	100	0.0560	11.6170
0.0	15.3846	200	0.0666	13.9717
0.0	23.0769	300	0.0754	13.8148
0.0	30.7692	400	0.0850	14.1287
0.0	38.4615	500	0.0945	13.1868
0.0	46.1538	600	0.1042	13.0298
0.0	53.8462	700	0.1147	13.8148
0.0	61.5385	800	0.1256	14.1287
0.0	69.2308	900	0.1361	14.7567
0.0	76.9231	1000	0.1487	13.8148
0.0	84.6154	1100	0.1619	17.1115
0.0	92.3077	1200	0.1759	17.2684
0.0	100.0	1300	0.1866	17.1115
0.0	107.6923	1400	0.1979	17.1115
0.0043	115.3846	1500	0.0933	10.2041
0.0	123.0769	1600	0.0901	10.9890
0.0	130.7692	1700	0.0914	11.3030
0.0	138.4615	1800	0.0922	11.3030
0.0	146.1538	1900	0.0929	11.3030
0.0	153.8462	2000	0.0933	11.3030
0.0	161.5385	2100	0.0938	11.3030
0.0	169.2308	2200	0.0942	11.3030
0.0	176.9231	2300	0.0943	11.3030
0.0	184.6154	2400	0.0945	11.3030
0.0	192.3077	2500	0.0947	11.3030
0.0	200.0	2600	0.0947	11.3030

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1