whisper-small-mn-3

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: 0.3277
• Wer: 30.3692
• Cer: 10.9030

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: 32
• eval_batch_size: 32
• 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: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3408	0.61	1000	0.4062	47.6841	17.3811
0.2261	1.22	2000	0.3262	37.8086	13.6466
0.2135	1.83	3000	0.2863	33.7175	12.2246
0.1643	2.43	4000	0.2803	32.5978	11.4526
0.1198	3.04	5000	0.2747	31.1121	11.0533
0.1279	3.65	6000	0.2757	30.7243	10.8927
0.0891	4.26	7000	0.2878	30.9209	11.0610
0.0899	4.87	8000	0.2906	30.6642	11.0799
0.0648	5.48	9000	0.3054	30.5986	10.9030
0.0436	6.09	10000	0.3184	30.5222	10.9434
0.0468	6.7	11000	0.3277	30.3692	10.9030
0.0291	7.3	12000	0.3411	30.9810	11.1572
0.0275	7.91	13000	0.3476	31.0684	11.1555
0.0196	8.52	14000	0.3572	30.9154	11.1065
0.0159	9.13	15000	0.3600	31.0356	11.2087

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2