Makkoen/whisper-large-v3-cit-do015-wd0-lr3e-06-FULL2

./2623

This model is a fine-tuned version of openai/whisper-large-v3 on the 2623 FULL dataset. It achieves the following results on the evaluation set:

• Loss: 0.5176
• Wer Ortho: 27.7646
• Wer: 19.7671

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• 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: 200
• training_steps: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.8323	0.6780	100	0.5981	33.0631	24.2505
0.5649	1.3559	200	0.5405	32.6716	24.1167
0.4921	2.0339	300	0.5132	30.1272	22.0155
0.3926	2.7119	400	0.5088	28.8271	21.1724
0.348	3.3898	500	0.5122	27.7925	19.6868
0.3125	4.0678	600	0.5093	28.2958	20.3560
0.2761	4.7458	700	0.5146	27.5828	19.5262
0.2664	5.4237	800	0.5176	27.7646	19.7671

Framework versions

• Transformers 4.44.0
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1