scream_non_large_timestamp_test

This model is a fine-tuned version of NbAiLab/scream_non_large_1e06_beams5_constantlr_long on the NbAiLab/NCC_whisper_both_timestamp dataset.

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-06
• lr_scheduler_type: linear
• per_device_train_batch_size: 16
• total_train_batch_size_per_node: 64
• total_train_batch_size: 64
• total_optimization_steps: 5,000
• starting_optimization_step: None
• finishing_optimization_step: 5,000
• num_train_dataset_workers: 32
• num_hosts: 1
• total_num_training_examples: 320,000
• steps_per_epoch: 475
• num_beams: 5

Training results

step	eval_loss	train_loss	eval_wer	eval_cer
0	0.5499	0.6409	10.5055	5.6538
500	0.4301	0.1060	10.1401	4.9232
1000	0.4168	0.0776	10.3837	4.8123
1500	0.4132	0.0689	10.5968	4.8123
2000	0.4176	0.0511	10.5968	4.6712
2500	0.4261	0.0454	11.0536	4.7871
3000	0.4385	0.0441	11.1449	4.7770

Framework versions

• Transformers 4.29.0.dev0
• Datasets 2.12.0
• Tokenizers 0.13.3