./openai/whisper-large-v3-cit-do015-wd0-lr5e-06-1000

This model is a fine-tuned version of openai/whisper-large-v3 on the SF 1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4753
• Wer Ortho: 23.5867
• Wer: 12.4052

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: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
No log	0.4444	25	1.1494	33.5283	21.6616
1.2689	0.8889	50	0.6362	28.0702	14.9090
1.2689	1.3333	75	0.5078	24.3275	12.2534
0.5452	1.7778	100	0.3860	23.1189	11.7602
0.5452	2.2222	125	0.3789	23.1969	11.1912
0.3251	2.6667	150	0.3691	24.0546	11.4568
0.3251	3.1111	175	0.3545	23.9376	11.5706
0.2441	3.5556	200	0.3701	25.3411	13.2018
0.2441	4.0	225	0.3564	21.4815	9.9393
0.1651	4.4444	250	0.3909	22.5731	10.3566
0.1651	4.8889	275	0.3708	24.6394	13.0121
0.1394	5.3333	300	0.3928	24.7563	13.2018
0.1394	5.7778	325	0.4097	24.6784	13.2018
0.1062	6.2222	350	0.4270	25.3021	13.4294
0.1062	6.6667	375	0.4133	24.2105	12.8225
0.0831	7.1111	400	0.4275	23.9766	13.0880
0.0831	7.5556	425	0.4592	23.1579	12.3293
0.065	8.0	450	0.4617	23.9376	12.5190
0.065	8.4444	475	0.4685	23.5088	12.4810
0.0558	8.8889	500	0.4753	23.5867	12.4052

Framework versions

• Transformers 4.42.3
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1