whisper-small-ko-EYfreqspeed

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

• Loss: 0.2179
• Cer: 21.0586

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.4086	0.3	100	0.2539	6.5745
0.3177	0.59	200	0.2342	6.3728
0.2886	0.89	300	0.2193	5.7557
0.1251	1.19	400	0.2201	5.9515
0.1171	1.48	500	0.2302	6.7762
0.1033	1.78	600	0.2181	5.6429
0.0363	2.07	700	0.2154	5.7616
0.039	2.37	800	0.2180	7.3874
0.0399	2.67	900	0.2113	17.4094
0.0367	2.96	1000	0.2080	14.2645
0.0146	3.26	1100	0.2151	16.5074
0.0128	3.56	1200	0.2155	20.8331
0.013	3.85	1300	0.2159	12.1225
0.0049	4.15	1400	0.2159	18.2757
0.0043	4.44	1500	0.2166	22.0970
0.0037	4.74	1600	0.2179	21.0586

Framework versions

• Transformers 4.34.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3