whisper_large_v2_thi_dataset_phase3

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4115
• Cer: 10.8763
• Wer: 20.7305

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
0.7544	0.9998	3132	17.4841	0.4112	30.6117
0.4151	2.0	6265	20.9872	0.3786	35.5513
0.3044	2.9995	9396	15.1611	0.3682	26.8405
0.3174	3.9998	12528	16.9391	0.3927	29.9976
0.2516	5.0	15661	14.2234	0.3991	25.0395
0.1993	5.9998	18793	12.4403	0.3980	22.7472
0.1597	7.0	21926	12.0030	0.4012	22.7070
0.1287	7.9998	25058	11.3252	0.4060	21.4210
0.1068	9.0	28191	11.0928	0.4107	21.1000
0.0916	9.9989	31320	10.8763	0.4115	20.7305
0.1047	10.9998	34452	0.4228	11.4684	21.5799
0.0896	12.0	37585	0.4303	10.9857	20.9462
0.076	12.9998	40717	0.4300	11.0458	20.9101
0.0651	14.0	43850	0.4352	10.8999	20.6779
0.0578	14.9992	46980	0.4382	10.9414	20.7904

Framework versions

• Transformers 4.41.2
• Pytorch 2.1.2+cu118
• Datasets 2.19.0
• Tokenizers 0.19.1