arun100/whisper-base-ko-1

Whisper Base Korean

This model is a fine-tuned version of openai/whisper-base on the mozilla-foundation/common_voice_16_0 ko dataset. It achieves the following results on the evaluation set:

• Loss: 0.6687
• Wer: 45.5026

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-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0149	133.0	1000	0.6687	45.5026
0.0048	266.0	2000	0.7148	47.7633
0.0024	399.0	3000	0.7484	48.4848
0.0014	533.0	4000	0.7774	49.0139
0.0009	666.0	5000	0.8037	48.8215
0.0006	799.0	6000	0.8269	49.4468
0.0004	933.0	7000	0.8482	49.3987
0.0003	1066.0	8000	0.8662	54.6417
0.0003	1199.0	9000	0.8800	49.9278
0.0003	1333.0	10000	0.8856	49.8316

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0