./openai/whisper-large-v3-cit-do015-wd0-lr1e-06-BALANCED

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

• Loss: 0.6001
• Wer Ortho: 32.5152
• Wer: 23.0724

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-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: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.1025	0.8368	50	0.8647	37.5455	28.4260
0.9017	1.6736	100	0.7168	37.0	29.4443
0.7253	2.5105	150	0.6533	34.0606	25.3710
0.681	3.3473	200	0.6284	38.5758	30.9281
0.6067	4.1841	250	0.6172	34.0909	26.3311
0.5794	5.0209	300	0.6089	34.0909	26.2438
0.5387	5.8577	350	0.6064	33.7576	25.9529
0.5171	6.6946	400	0.6025	32.7273	23.2179
0.5322	7.5314	450	0.6006	36.0909	26.1856
0.5069	8.3682	500	0.6001	32.5152	23.0724

Framework versions

• Transformers 4.42.4
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1