GGarri/241002_whisperfinetuned

Whisper Small ko

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

• Loss: 0.0268
• Cer: 6.5045
• Wer: 6.9310

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: 32
• 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: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
3.6391	0.54	25	3.3230	83.6552	35.4340
2.7648	1.09	50	2.3011	81.2725	31.6473
1.8272	1.63	75	1.4490	85.9460	43.8688
1.0827	2.17	100	0.8137	72.8033	59.1524
0.6201	2.72	125	0.4756	50.5476	49.9522
0.3539	3.26	150	0.3005	31.1094	31.5926
0.2358	3.8	175	0.1969	29.5962	31.3192
0.1501	4.35	200	0.1352	21.1688	21.7772
0.0967	4.89	225	0.0846	18.6941	19.0431
0.0471	5.43	250	0.0350	18.3931	18.9200
0.0162	5.98	275	0.0335	18.9616	19.5215
0.0121	6.52	300	0.0324	14.1293	15.5707
0.011	7.07	325	0.0261	12.9755	14.3267
0.0078	7.61	350	0.0223	9.3220	10.5400
0.0075	8.15	375	0.0217	5.8106	6.5482
0.0052	8.7	400	0.0208	7.9926	8.6945
0.0048	9.24	425	0.0213	5.3424	5.7280
0.0053	9.78	450	0.0212	7.5328	7.9973
0.004	10.33	475	0.0213	5.7186	5.9740
0.0054	10.87	500	0.0268	6.5045	6.9310

Framework versions

• Transformers 4.39.2
• Pytorch 2.0.1
• Datasets 2.18.0
• Tokenizers 0.15.2