whisper-tiny-polyai-enUS_lower_lr

This model is a fine-tuned version of openai/whisper-tiny on the PolyAI/minds14 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7730
• Wer Ortho: 0.4022
• Wer: 0.3849

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-06
• 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: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
2.9571	3.33	50	1.9622	0.5077	0.4050
0.5131	6.67	100	0.6540	0.4152	0.3684
0.2572	10.0	150	0.6091	0.3874	0.3524
0.0974	13.33	200	0.6316	0.3745	0.3442
0.0405	16.67	250	0.6686	0.3917	0.3577
0.0116	20.0	300	0.7097	0.4028	0.3766
0.0049	23.33	350	0.7341	0.3954	0.3743
0.0032	26.67	400	0.7510	0.4065	0.3884
0.0023	30.0	450	0.7607	0.3967	0.3778
0.0018	33.33	500	0.7730	0.4022	0.3849

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0