Hiro-nguyen/whisper-small-hi

Whisper Base EN

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

• Loss: 0.0017
• Wer: 1.3384

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: 8
• 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: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.7899	4.1667	100	2.3530	21.4149
0.8377	8.3333	200	0.7500	4.2065
0.0599	12.5	300	0.0394	1.9120
0.0163	16.6667	400	0.0151	2.1033
0.0068	20.8333	500	0.0023	1.1472
0.0031	25.0	600	0.0018	1.3384
0.0027	29.1667	700	0.0023	1.3384
0.0018	33.3333	800	0.0020	1.3384
0.003	37.5	900	0.0017	1.3384
0.0009	41.6667	1000	0.0017	1.3384

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0a0+ebedce2
• Datasets 2.19.2
• Tokenizers 0.19.1