finetune_v15

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7837
• Wer: 193.6017

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-05
• train_batch_size: 16
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5
• training_steps: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	6.1538	10	0.7300	34.1589
No log	12.3077	20	0.7090	39.9381
No log	18.4615	30	0.7617	33.2559
No log	24.6154	40	0.7676	33.4107
0.223	30.7692	50	0.7749	199.6646
0.223	36.9231	60	0.7764	164.3189
0.223	43.0769	70	0.7827	202.6574
0.223	49.2308	80	0.7837	193.6017

Framework versions

• Transformers 4.42.4
• Pytorch 2.2.0
• Datasets 2.20.0
• Tokenizers 0.19.1