metadata

language:
  - ko
license: apache-2.0
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
base_model: openai/whisper-small
datasets:
  - korean_samll_dataset4
model-index:
  - name: korean-small_t35
    results: []

korean-small_t35

This model is a fine-tuned version of openai/whisper-small on the korean_samll_dataset4 dataset. It achieves the following results on the evaluation set:

Loss: 0.1385
Cer: 5.2553

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
num_epochs: 3.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.2501	0.06	200	0.2322	8.7579
0.217	0.12	400	0.2118	8.6072
0.1947	0.18	600	0.2011	7.4035
0.1938	0.24	800	0.1941	7.3665
0.1878	0.3	1000	0.1826	7.0860
0.192	0.36	1200	0.1786	6.8894
0.1768	0.42	1400	0.1739	6.5072
0.1777	0.48	1600	0.1708	6.4205
0.1714	0.54	1800	0.1675	6.6288
0.171	0.6	2000	0.1637	6.3026
0.1678	0.66	2200	0.1638	6.4964
0.1606	0.72	2400	0.1604	6.4205
0.1541	0.78	2600	0.1580	6.1524
0.1578	0.84	2800	0.1550	5.8736
0.1524	0.9	3000	0.1535	5.9458
0.153	0.96	3200	0.1512	5.8205
0.112	1.02	3400	0.1492	5.7590
0.0833	1.08	3600	0.1491	5.7022
0.0928	1.14	3800	0.1495	5.6578
0.1005	1.2	4000	0.1480	6.0906
0.0918	1.26	4200	0.1475	5.8175
0.0929	1.32	4400	0.1470	5.7632
0.091	1.38	4600	0.1460	5.6557
0.0858	1.44	4800	0.1445	5.6947
0.0889	1.5	5000	0.1435	5.6632
0.0903	1.56	5200	0.1442	5.6412
0.0894	1.61	5400	0.1426	5.5711
0.0842	1.67	5600	0.1426	5.4424
0.0926	1.73	5800	0.1419	5.4171
0.0801	1.79	6000	0.1400	5.3960
0.0843	1.85	6200	0.1397	5.5648
0.0909	1.91	6400	0.1386	5.4677
0.0816	1.97	6600	0.1384	5.6586
0.0484	2.03	6800	0.1421	5.4541
0.0506	2.09	7000	0.1408	5.4424
0.0475	2.15	7200	0.1410	5.5565
0.0477	2.21	7400	0.1406	5.5453
0.0465	2.27	7600	0.1407	5.3383
0.0487	2.33	7800	0.1404	5.4192
0.0438	2.39	8000	0.1400	5.4088
0.0432	2.45	8200	0.1404	5.4022
0.0457	2.51	8400	0.1410	5.3852
0.0468	2.57	8600	0.1398	5.2881
0.0456	2.63	8800	0.1390	5.2789
0.0426	2.69	9000	0.1390	5.3619
0.0437	2.75	9200	0.1385	5.2553
0.0467	2.81	9400	0.1386	5.3404
0.044	2.87	9600	0.1383	5.3180
0.0445	2.93	9800	0.1382	5.3072
0.0453	2.99	10000	0.1380	5.3267

Framework versions

Transformers 4.39.0.dev0
Pytorch 2.2.1+cu121
Datasets 2.18.0
Tokenizers 0.15.2