NbAiLab/whisper-tiny-gputest

Whisper Tiny GPU test

This model is a fine-tuned version of openai/whisper-tiny on the NbAiLab/NCC_S3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.9375
• Wer: 51.3703

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: 3e-06
• train_batch_size: 128
• 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_steps: 200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.4574	0.1	200	1.4663	71.6504
1.9587	0.2	400	1.2581	64.7381
1.816	0.3	600	1.1672	60.9318
1.7199	0.4	800	1.1006	57.6736
1.6686	0.5	1000	1.0630	56.1815
1.621	0.6	1200	1.0273	55.4811
1.5846	0.7	1400	1.0017	53.9890
1.5482	0.8	1600	0.9773	53.0146
1.521	0.9	1800	0.9575	52.1011
1.4932	1.0	2000	0.9375	51.3703

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2