whisper-large-cit-do1.5-wd1e-3-lr5

This model is a fine-tuned version of openai/whisper-large-v3 on the SF 200 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8623
• Wer: 27.9176

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: 5e-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9999	0.8889	10	0.8228	34.3249
0.7031	1.7778	20	0.6328	32.2654
0.4625	2.6667	30	0.5498	30.4348
0.2785	3.5556	40	0.5278	32.2654
0.1827	4.4444	50	0.5557	28.6041
0.1029	5.3333	60	0.6138	28.3753
0.06	6.2222	70	0.6641	29.7483
0.0266	7.1111	80	0.7666	29.0618
0.0229	8.0	90	0.7114	29.9771
0.0143	8.8889	100	0.7417	27.0023
0.0183	9.7778	110	0.8423	30.8924
0.0115	10.6667	120	0.7061	29.0618
0.0091	11.5556	130	0.7661	28.8330
0.0029	12.4444	140	0.8232	28.1465
0.0064	13.3333	150	0.8213	29.5195
0.0032	14.2222	160	0.8389	27.6888
0.0021	15.1111	170	0.8511	28.3753
0.0023	16.0	180	0.8545	28.3753
0.0015	16.8889	190	0.8599	28.1465
0.0013	17.7778	200	0.8623	27.9176

Framework versions

• Transformers 4.41.1
• Pytorch 1.13.1+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1