Makkoen/whisper-large-v3-cit-do005-wd0-lr3e-06-steps1400-FULL5

./7326

This model is a fine-tuned version of openai/whisper-large-v3 on the 7326 FULL-2024-10-24 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3906
• Wer Ortho: 22.5859
• Wer: 15.5145

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: 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: 300
• training_steps: 1400
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.6841	0.4851	200	0.4590	25.7036	18.3842
0.5245	0.9703	400	0.4204	24.1509	16.9945
0.4305	1.4554	600	0.4028	23.1144	15.9502
0.4039	1.9406	800	0.3940	23.1431	16.0444
0.3567	2.4257	1000	0.3943	22.6269	15.6754
0.3391	2.9109	1200	0.3904	22.5900	15.5459
0.317	3.3960	1400	0.3906	22.5859	15.5145

Framework versions

• Transformers 4.45.1
• Pytorch 1.13.1+cu117
• Datasets 3.0.1
• Tokenizers 0.20.0