README.md

metadata

library_name: transformers
language:
  - ko
license: apache-2.0
base_model: openai/whisper-small
tags:
  - whisper-event
  - generated_from_trainer
datasets:
  - GGarri/241109_newdata
metrics:
  - wer
model-index:
  - name: Whisper Small ko
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: customdata
          type: GGarri/241109_newdata
        metrics:
          - name: Wer
            type: wer
            value: 1.3576367064739157

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: 0.0892
• Cer: 1.7222
• Wer: 1.3576

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: 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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.0297	1.9231	100	0.8542	5.7071	5.2420
0.2945	3.8462	200	0.2539	3.1678	2.6524
0.022	5.7692	300	0.0721	2.3633	1.8982
0.0066	7.6923	400	0.0744	2.1999	1.7348
0.0055	9.6154	500	0.0681	1.9485	1.5462
0.0021	11.5385	600	0.0743	2.1622	1.7096
0.0019	13.4615	700	0.0723	2.1747	1.7096
0.0005	15.3846	800	0.0733	1.8856	1.4833
0.0001	17.3077	900	0.0738	1.9233	1.4582
0.0001	19.2308	1000	0.0748	1.9233	1.4582
0.0001	21.1538	1100	0.0759	1.8353	1.4708
0.0001	23.0769	1200	0.0765	1.8102	1.4456
0.0001	25.0	1300	0.0770	1.7976	1.4331
0.0001	26.9231	1400	0.0773	1.7976	1.4331
0.0001	28.8462	1500	0.0776	1.7976	1.4331
0.0001	30.7692	1600	0.0780	1.7976	1.4331
0.0001	32.6923	1700	0.0782	1.7976	1.4331
0.0001	34.6154	1800	0.0786	1.7850	1.4205
0.0001	36.5385	1900	0.0790	1.7850	1.4205
0.0001	38.4615	2000	0.0794	1.7599	1.3953
0.0001	40.3846	2100	0.0804	1.7599	1.3953
0.0001	42.3077	2200	0.0811	1.7599	1.3953
0.0	44.2308	2300	0.0816	1.7599	1.3953
0.0	46.1538	2400	0.0821	1.7473	1.3828
0.0	48.0769	2500	0.0825	1.7473	1.3828
0.0	50.0	2600	0.0829	1.7473	1.3828
0.0	51.9231	2700	0.0832	1.7473	1.3828
0.0	53.8462	2800	0.0836	1.7473	1.3828
0.0	55.7692	2900	0.0840	1.7473	1.3828
0.0	57.6923	3000	0.0843	1.7473	1.3828
0.0	59.6154	3100	0.0846	1.7473	1.3828
0.0	61.5385	3200	0.0849	1.7473	1.3828
0.0	63.4615	3300	0.0853	1.7473	1.3828
0.0	65.3846	3400	0.0855	1.7348	1.3702
0.0	67.3077	3500	0.0858	1.7348	1.3702
0.0	69.2308	3600	0.0860	1.7348	1.3702
0.0	71.1538	3700	0.0863	1.7348	1.3702
0.0	73.0769	3800	0.0866	1.7348	1.3702
0.0	75.0	3900	0.0868	1.7348	1.3702
0.0	76.9231	4000	0.0870	1.7348	1.3702
0.0	78.8462	4100	0.0877	1.7473	1.3828
0.0	80.7692	4200	0.0881	1.7222	1.3576
0.0	82.6923	4300	0.0884	1.7222	1.3576
0.0	84.6154	4400	0.0886	1.7222	1.3576
0.0	86.5385	4500	0.0888	1.7222	1.3576
0.0	88.4615	4600	0.0889	1.7222	1.3576
0.0	90.3846	4700	0.0890	1.7222	1.3576
0.0	92.3077	4800	0.0891	1.7222	1.3576
0.0	94.2308	4900	0.0891	1.7222	1.3576
0.0	96.1538	5000	0.0892	1.7222	1.3576

Framework versions

• Transformers 4.46.2
• Pytorch 2.4.0
• Datasets 2.18.0
• Tokenizers 0.20.3