gwanju4_test__model

This model is a fine-tuned version of openai/whisper-medium on the Marcusxx/gwanju4 dataset. It achieves the following results on the evaluation set:

• Loss: 2.5722
• Cer: 275.0

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
1.9955	50.0	50	4.8936	425.0
0.3976	100.0	100	2.6194	275.0
0.0001	150.0	150	2.5074	275.0
0.0	200.0	200	2.4715	325.0
0.0	250.0	250	2.4384	325.0
0.0	300.0	300	2.3972	275.0
0.0	350.0	350	2.3937	275.0
0.0	400.0	400	2.4084	275.0
0.0	450.0	450	2.4310	275.0
0.0	500.0	500	2.4561	275.0
0.0	550.0	550	2.4782	275.0
0.0	600.0	600	2.5006	275.0
0.0	650.0	650	2.5150	275.0
0.0	700.0	700	2.5310	275.0
0.0	750.0	750	2.5415	275.0
0.0	800.0	800	2.5517	275.0
0.0	850.0	850	2.5592	275.0
0.0	900.0	900	2.5674	275.0
0.0	950.0	950	2.5723	275.0
0.0	1000.0	1000	2.5722	275.0

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1