Makkoen/whisper-large-cit-do0.2-wd0.001-lr5

whisper-large-cit-do0.2-wd0.001-tr5

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: 0.8701
• Wer: 28.8330

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
1.0205	0.8889	10	0.8452	34.3249
0.7299	1.7778	20	0.6455	31.1213
0.48	2.6667	30	0.5552	30.4348
0.291	3.5556	40	0.5288	30.6636
0.1931	4.4444	50	0.5479	28.3753
0.107	5.3333	60	0.6104	29.0618
0.0622	6.2222	70	0.6509	28.8330
0.0271	7.1111	80	0.7900	30.4348
0.0198	8.0	90	0.7246	30.2059
0.0176	8.8889	100	0.6992	27.6888
0.0163	9.7778	110	0.7896	29.5195
0.0087	10.6667	120	0.7793	30.4348
0.0092	11.5556	130	0.8213	28.8330
0.0063	12.4444	140	0.8369	29.5195
0.0038	13.3333	150	0.8262	29.7483
0.0036	14.2222	160	0.8506	28.3753
0.0021	15.1111	170	0.8647	29.5195
0.0017	16.0	180	0.8608	29.5195
0.0012	16.8889	190	0.8662	28.8330
0.001	17.7778	200	0.8701	28.8330

Framework versions

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