whisper-training-blog

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

• Loss: 1.0050
• Wer: 191.2342

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: 7.5e-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.3
• training_steps: 448
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4111	0.1	44	1.4919	245.3457
1.0501	0.2	88	1.2255	225.8822
0.9032	0.29	132	1.1203	211.6558
0.8141	1.06	176	1.0675	184.6240
0.8029	1.16	220	1.0394	178.4129
0.6325	1.25	264	1.0301	216.6374
0.6971	2.02	308	1.0135	184.4004
0.6051	2.12	352	1.0065	194.7150
0.6047	2.21	396	1.0029	166.9328
0.585	2.31	440	1.0050	191.2342

Framework versions

• Transformers 4.29.2
• Pytorch 2.0.1+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3