chungnam_firestation3Kfiles_WER_model

This model is a fine-tuned version of openai/whisper-medium on the Marcusxx/chungnamFireStation3Kfiles dataset. It achieves the following results on the evaluation set:

• Loss: 1.2164
• Wer: 54.3478

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: 2000
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0826	6.6667	1000	0.6837	170.3934
0.0377	13.3333	2000	0.7952	108.2816
0.0131	20.0	3000	0.8730	56.7288
0.0066	26.6667	4000	0.8840	54.6584
0.0015	33.3333	5000	0.9585	55.2795
0.0123	40.0	6000	0.9682	55.9006
0.0006	46.6667	7000	1.0101	53.7267
0.0022	53.3333	8000	1.0249	55.7971
0.002	60.0	9000	1.0083	57.4534
0.0	66.6667	10000	1.0515	55.3830
0.0	73.3333	11000	1.0877	54.7619
0.0	80.0	12000	1.1063	54.9689
0.0	86.6667	13000	1.1240	55.3830
0.0	93.3333	14000	1.1405	55.1760
0.0	100.0	15000	1.1579	54.7619
0.0	106.6667	16000	1.1736	54.9689
0.0	113.3333	17000	1.1890	54.9689
0.0	120.0	18000	1.2021	54.4513
0.0	126.6667	19000	1.2120	54.3478
0.0	133.3333	20000	1.2164	54.3478

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1