NbAiLab/whisper-small-smj-test

whisper-small-smj

This model is a fine-tuned version of openai/whisper-small on the NbAiLab/salmon-asr-smj dataset. It achieves the following results on the evaluation set:

• step: 9999
• validation_loss: 0.3690
• train_loss: 0.2159
• validation_wer: 19.6809
• validation_cer: 5.5037
• validation_exact_wer: 22.3404
• validation_exact_cer: 5.8753

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
• lr_scheduler_type: linear
• per_device_train_batch_size: 32
• total_train_batch_size_per_node: 256
• total_train_batch_size: 256
• total_optimization_steps: 10,000
• starting_optimization_step: None
• finishing_optimization_step: 10,000
• num_train_dataset_workers: 32
• num_hosts: 1
• total_num_training_examples: 2,560,000
• steps_per_epoch: 70
• num_beams: None
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.98
• adam_epsilon: 1e-06
• dropout: True
• bpe_dropout_probability: 0.2
• activation_dropout_probability: 0.1

Training results

step	validation_loss	train_loss	validation_wer	validation_cer	validation_exact_wer	validation_exact_cer
0	3.4458	4.7979	205.7181	94.0902	150.2660	95.4591
1000	0.8415	0.2440	21.9415	6.4379	25.9309	7.0106
2000	1.0741	0.2249	21.6755	5.7474	25.0	6.1741
3000	0.8933	0.2919	20.4787	5.3615	23.9362	5.8156
4000	0.8445	0.1339	18.8830	5.2193	21.4096	5.6363
5000	0.3739	0.2289	20.0798	5.3818	23.2713	5.8355
6000	0.3746	0.2586	19.8138	5.2600	22.7394	5.6562
7000	0.3555	0.2273	19.2819	5.7067	22.3404	6.0745
8000	0.3671	0.1632	19.4149	5.4224	22.3404	5.8952
9000	0.3508	0.2107	18.3511	5.3006	21.2766	5.7160
9999	0.3690	0.2159	19.6809	5.5037	22.3404	5.8753

Framework versions

• Transformers 4.34.1
• Datasets 2.14.5
• Tokenizers 0.14.1