JunWorks/whisperBase_LoRA_zhTW

Whisper Base LoRA tuned zh-TW

DEMO LINK: https://4e766dca651b881c9b.gradio.live

This model is a fine-tuned version with PEFT-LoRA of openai/whisper-base on the Common Voice 13.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3839
• CER: 22.13% :(

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: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.3981	1.0	1453	0.4017
0.4016	2.0	2906	0.3960
0.3543	3.0	4359	0.3933
0.3638	4.0	5812	0.3905
0.3953	5.0	7265	0.3895
0.377	6.0	8718	0.3879
0.3646	7.0	10171	0.3869
0.3592	8.0	11624	0.3860
0.3324	9.0	13077	0.3853
0.3818	10.0	14530	0.3848
0.3107	11.0	15983	0.3844
0.3473	12.0	17436	0.3844
0.3684	13.0	18889	0.3845
0.3886	14.0	20342	0.3841
0.3652	15.0	21795	0.3839

Framework versions

• PEFT 0.9.1.dev0
• Transformers 4.39.0.dev0
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2