whisper-tiny-us_en_bs128

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: 0.8372
• Wer Ortho: 0.3399
• Wer: 0.3418

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: 0.0001
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.1633	6.25	25	0.5503	0.3177	0.3164
0.0027	12.5	50	0.6676	0.3288	0.3294
0.0011	18.75	75	0.7095	0.3134	0.3182
0.0012	25.0	100	0.7296	0.3196	0.3176
0.0014	31.25	125	0.7460	0.3541	0.3583
0.005	37.5	150	0.7059	0.4405	0.4610
0.0009	43.75	175	0.7803	0.3924	0.3961
0.0004	50.0	200	0.7996	0.3455	0.3512
0.0001	56.25	225	0.8074	0.3411	0.3442
0.0001	62.5	250	0.8146	0.3424	0.3459
0.0001	68.75	275	0.8197	0.3430	0.3459
0.0001	75.0	300	0.8239	0.3399	0.3424
0.0001	81.25	325	0.8274	0.3374	0.3400
0.0001	87.5	350	0.8303	0.3356	0.3383
0.0001	93.75	375	0.8324	0.3368	0.3400
0.0001	100.0	400	0.8341	0.3368	0.3388
0.0001	106.25	425	0.8354	0.3405	0.3424
0.0001	112.5	450	0.8364	0.3399	0.3418
0.0001	118.75	475	0.8371	0.3399	0.3418
0.0001	125.0	500	0.8372	0.3399	0.3418

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3