./openai/whisper-large-v3-cit-do015-wd0-lr3e-06-FULL

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: 0.5117
• Wer Ortho: 27.7362
• Wer: 18.6050

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: 100
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.9582	0.4773	50	0.6479	34.9922	25.6303
0.6764	0.9547	100	0.5605	30.9901	21.5126
0.5263	1.4320	150	0.5337	29.3892	20.0168
0.5084	1.9093	200	0.5186	28.0842	19.1261
0.4226	2.3866	250	0.5132	27.9624	18.8571
0.4078	2.8640	300	0.5083	28.1538	19.0420
0.3775	3.3413	350	0.5083	28.3974	18.8403
0.3532	3.8186	400	0.5093	28.1538	18.6555
0.3359	4.2959	450	0.5098	27.7188	18.5210
0.3189	4.7733	500	0.5117	27.7362	18.6050

Framework versions

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