cheonanAddresses_torch_medium_30000_model

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

• Loss: 0.0824
• Cer: 1.9162

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: 100
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.0752	0.3101	1000	0.0768	51.6768
0.0727	0.6202	2000	0.0740	27.4313
0.0635	0.9302	3000	0.0692	2.3148
0.0653	1.2403	4000	0.0660	2.3503
0.062	1.5504	5000	0.0645	2.2620
0.0595	1.8605	6000	0.0621	2.2244
0.0545	2.1705	7000	0.0595	2.2490
0.0515	2.4806	8000	0.0599	2.1043
0.0594	2.7907	9000	0.0573	2.0146
0.0428	3.1008	10000	0.0592	2.0609
0.05	3.4109	11000	0.0565	2.0645
0.0484	3.7209	12000	0.0541	1.9336
0.0367	4.0310	13000	0.0560	1.9727
0.0293	4.3411	14000	0.0589	1.8887
0.035	4.6512	15000	0.0541	1.9032
0.0313	4.9612	16000	0.0536	1.9003
0.0227	5.2713	17000	0.0636	1.9517
0.026	5.5814	18000	0.0617	1.9379
0.0268	5.8915	19000	0.0589	1.9104
0.017	6.2016	20000	0.0673	1.9596
0.0174	6.5116	21000	0.0666	1.9264
0.0114	6.8217	22000	0.0688	1.8844
0.0063	7.1318	23000	0.0733	1.9900
0.0047	7.4419	24000	0.0761	1.9669
0.0074	7.7519	25000	0.0764	1.9560
0.0024	8.0620	26000	0.0778	1.9821
0.002	8.3721	27000	0.0794	1.9307
0.0024	8.6822	28000	0.0811	1.9589
0.0019	8.9922	29000	0.0815	1.9198
0.001	9.3023	30000	0.0824	1.9162

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1