salmon-whisper-large-smj-lr7e-5-test1

This model is a fine-tuned version of openai/whisper-large-v2 on the NbAiLab/salmon-asr-smj dataset. It achieves the following results on the evaluation set:

• step: 999
• validation_loss: 0.9447
• train_loss: 0.3067
• validation_wer: 21.6755
• validation_cer: 5.6661
• validation_exact_wer: 25.0
• validation_exact_cer: 6.1940

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: 7e-05
• lr_scheduler_type: linear
• per_device_train_batch_size: 6
• total_train_batch_size_per_node: 48
• total_train_batch_size: 48
• total_optimization_steps: 1,000
• starting_optimization_step: None
• finishing_optimization_step: 1,000
• num_train_dataset_workers: 32
• num_hosts: 1
• total_num_training_examples: 48,000
• steps_per_epoch: 385
• 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	4.2254	4.6455	112.7660	59.8700	108.1117	62.0594
100	1.4819	0.9353	59.0426	16.0032	61.8351	16.8293
200	1.2494	0.8903	43.2181	10.9667	45.8777	11.6311
300	1.1444	0.8144	32.4468	8.4281	35.6383	8.8429
400	1.0442	1.3240	30.1862	7.7173	33.3777	8.2454
500	0.9681	0.2736	25.5319	6.3769	28.4574	6.8711
600	1.0579	0.4364	25.0	6.3363	28.3245	6.8313
700	0.9322	0.6873	23.4043	5.9708	26.3298	6.3732
800	0.9255	0.3675	23.2713	6.0114	26.5957	6.5326
900	0.9581	0.6156	22.4734	5.8692	26.0638	6.4330
999	0.9447	0.3067	21.6755	5.6661	25.0	6.1940

Framework versions

• Transformers 4.34.1
• Datasets 2.15.0
• Tokenizers 0.14.0