mvbnh/whisper-small-ko-new

Whisper Small Ko

This model is a fine-tuned version of openai/whisper-small on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0919
• Wer: 5.0143

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0956	0.0718	100	0.1442	11.5193
0.0808	0.1437	200	0.1401	10.9622
0.0621	0.2155	300	0.1385	10.7815
0.052	0.2874	400	0.1372	10.4803
0.0442	0.3592	500	0.1401	9.7124
0.0454	0.4310	600	0.1369	10.7213
0.0483	0.5029	700	0.1365	10.8568
0.0493	0.5747	800	0.1329	9.1251
0.0393	0.6466	900	0.1253	8.5379
0.0619	0.7184	1000	0.1227	7.7699
0.0863	0.7902	1100	0.1193	8.4325
0.0786	0.8621	1200	0.1165	8.4927
0.0825	0.9339	1300	0.1130	8.5379
0.0707	1.0057	1400	0.1085	7.3483
0.0254	1.0776	1500	0.1049	6.3996
0.0216	1.1494	1600	0.1043	6.2189
0.0204	1.2213	1700	0.1025	6.9568
0.0222	1.2931	1800	0.1025	6.2491
0.0231	1.3649	1900	0.1032	6.7761
0.0258	1.4368	2000	0.1004	7.1224
0.0228	1.5086	2100	0.1000	7.2429
0.0222	1.5805	2200	0.0995	5.8576
0.0197	1.6523	2300	0.0959	5.4811
0.0239	1.7241	2400	0.0952	6.9568
0.0223	1.7960	2500	0.0948	6.6255
0.0145	1.8678	2600	0.0931	5.7823
0.022	1.9397	2700	0.0906	5.4058
0.011	2.0115	2800	0.0916	5.5112
0.007	2.0833	2900	0.0921	5.2552
0.0065	2.1552	3000	0.0924	6.0533
0.0088	2.2270	3100	0.0928	5.3757
0.0063	2.2989	3200	0.0931	5.2552
0.0058	2.3707	3300	0.0922	5.2853
0.0086	2.4425	3400	0.0924	5.1950
0.0149	2.5144	3500	0.0918	5.3305
0.0087	2.5862	3600	0.0915	5.3456
0.0081	2.6580	3700	0.0918	5.2251
0.0083	2.7299	3800	0.0922	5.3305
0.0062	2.8017	3900	0.0920	5.0896
0.0077	2.8736	4000	0.0919	5.0143

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1