whisper-NST-cons2e5

This model is a fine-tuned version of openai/whisper-small on the NBAILAB/NST - NO-CLOSE dataset. It achieves the following results on the evaluation set:

• Loss: 0.3521
• Wer: 11.8586

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2517	0.1	1000	0.4131	18.4721
0.1931	0.2	2000	0.3531	19.0422
0.1598	0.3	3000	0.3605	16.8757
0.1541	0.4	4000	0.3367	14.4812
0.1443	0.5	5000	0.3274	13.3409
0.1301	0.6	6000	0.3481	10.7184
0.1266	0.7	7000	0.3452	12.9989
0.1216	0.8	8000	0.3215	10.8324
0.1121	0.9	9000	0.3160	11.5165
0.1171	1.0	10000	0.3521	11.8586

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.6.1
• Tokenizers 0.13.1