whisper-small-ko-baseline

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

• Loss: 0.2914
• Cer: 8.5820

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.2433	0.1289	100	0.2943	7.8712
0.1379	0.2579	200	0.2758	7.4248
0.1105	0.3868	300	0.2831	7.6010
0.1231	0.5158	400	0.2671	7.2192
0.0984	0.6447	500	0.2721	7.2603
0.0938	0.7737	600	0.2742	7.1840
0.0954	0.9026	700	0.2718	6.9901
0.0385	1.0316	800	0.2649	6.9549
0.0302	1.1605	900	0.2645	7.5129
0.0388	1.2895	1000	0.2722	7.0606
0.0338	1.4184	1100	0.2819	7.8889
0.0389	1.5474	1200	0.2725	7.7479
0.0335	1.6763	1300	0.2716	8.3647
0.0331	1.8053	1400	0.2751	7.6774
0.0343	1.9342	1500	0.2825	7.8008
0.0134	2.0632	1600	0.2739	6.9079
0.0127	2.1921	1700	0.2779	8.8287
0.0141	2.3211	1800	0.2822	7.1429
0.0113	2.4500	1900	0.2864	8.6407
0.0131	2.5790	2000	0.2797	10.5909
0.0103	2.7079	2100	0.2835	8.4880
0.0117	2.8369	2200	0.2828	11.5425
0.0116	2.9658	2300	0.2832	9.5747
0.0046	3.0948	2400	0.2862	8.8640
0.0045	3.2237	2500	0.2877	10.0388
0.0061	3.3527	2600	0.2886	8.9991
0.0055	3.4816	2700	0.2894	8.4704
0.0052	3.6106	2800	0.2904	8.4410
0.0059	3.7395	2900	0.2908	10.3266
0.0051	3.8685	3000	0.2913	9.3280
0.0047	3.9974	3100	0.2914	8.5820

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0