gwanju_small2_model

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

• Loss: 0.5890
• Cer: 213.0841

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
• training_steps: 25000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.5224	0.2964	1000	0.5003	164.2695
0.4399	0.5928	2000	0.4550	440.4495
0.4331	0.8892	3000	0.4277	151.5039
0.2926	1.1855	4000	0.4221	83.3523
0.3307	1.4819	5000	0.4162	221.7744
0.2858	1.7783	6000	0.4099	231.1391
0.1883	2.0747	7000	0.4122	71.8191
0.1794	2.3711	8000	0.4186	64.8286
0.1957	2.6675	9000	0.4146	147.6546
0.1752	2.9638	10000	0.4173	90.3213
0.1158	3.2602	11000	0.4346	187.8218
0.1216	3.5566	12000	0.4342	112.4640
0.107	3.8530	13000	0.4401	101.8964
0.0679	4.1494	14000	0.4593	153.6522
0.0647	4.4458	15000	0.4712	91.6056
0.0646	4.7421	16000	0.4732	97.5377
0.0289	5.0385	17000	0.4958	170.3914
0.0343	5.3349	18000	0.5112	160.5715
0.0319	5.6313	19000	0.5129	147.4355
0.0344	5.9277	20000	0.5226	129.4900
0.018	6.2241	21000	0.5561	181.4288
0.0179	6.5205	22000	0.5620	191.2249
0.0177	6.8168	23000	0.5668	190.5664
0.0086	7.1132	24000	0.5841	210.2615
0.0085	7.4096	25000	0.5890	213.0841

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1