whisper-base-full-data-v2

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

• Loss: 0.1481

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: tpu
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 256
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5000
• training_steps: 63840

Training results

Training Loss	Epoch	Step	Validation Loss
0.2986	1.57	5000	0.3998
0.2037	3.13	10000	0.3051
0.1683	4.7	15000	0.2646
0.1426	6.27	20000	0.2384
0.1265	7.83	25000	0.2186
0.1043	9.4	30000	0.2013
0.0971	10.97	35000	0.1894
0.0801	12.53	40000	0.1791
0.0654	14.1	45000	0.1703
0.0583	15.67	50000	0.1614
0.0471	17.23	55000	0.1553
0.0411	18.8	60000	0.1501

Framework versions

• Transformers 4.28.1
• Pytorch 2.1.0a0+gitcc01568
• Datasets 2.12.0
• Tokenizers 0.13.3