trained_model

This model is a fine-tuned version of byoussef/whisper-large-v2-Ko on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5559
• Cer: 27.1916

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: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 17460
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
1.2023	0.44	1000	61.2630	1.0742
1.1553	0.89	2000	48.5561	0.9965
0.5591	1.33	3000	41.7970	1.0181
0.4999	1.78	4000	42.4589	0.9979
0.3963	2.22	5000	56.0316	1.0151
0.1948	2.67	6000	39.5938	1.0179
0.8484	3.11	7000	42.2062	0.8630
0.4461	3.56	8000	41.4453	0.8244
0.5309	4.0	9000	38.5052	0.7959
0.4761	4.44	10000	0.5227	31.4439
0.5826	4.89	11000	0.5017	32.8738
0.2989	5.33	12000	0.5057	29.9159
0.2387	5.78	13000	0.5043	30.0935
0.079	6.22	14000	0.5264	35.1729
0.19	6.67	15000	0.5234	27.0327
0.1049	7.11	16000	0.5564	27.4953
0.0449	7.56	17000	0.5559	27.1916

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0