whisper-medium-pt-nonverbal-v3-1-demo

This model is a fine-tuned version of openai/whisper-medium on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5803
• Wer: 47.7893

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.3675	0.2841	100	2.3957	136.7360
0.3724	0.5682	200	0.7022	141.4824
0.2198	0.8523	300	0.6064	113.4590
0.1343	1.1364	400	0.6187	68.3355
0.1647	1.4205	500	0.6852	88.4915
0.1516	1.7045	600	0.6420	230.7542
0.1679	1.9886	700	0.4757	58.0624
0.0954	2.2727	800	0.5675	48.1144
0.0653	2.5568	900	0.7046	66.2549
0.0906	2.8409	1000	0.4930	41.6775
0.0363	3.125	1100	0.5209	54.2913
0.0476	3.4091	1200	0.5836	49.6099
0.075	3.6932	1300	0.5253	49.1547
0.0274	3.9773	1400	0.5842	48.3095
0.0118	4.2614	1500	0.5568	45.0585
0.0121	4.5455	1600	0.5740	48.6996
0.0111	4.8295	1700	0.5486	48.8296
0.0055	5.1136	1800	0.5646	46.8791
0.0032	5.3977	1900	0.5771	47.9844
0.0043	5.6818	2000	0.5803	47.7893

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1