chengyili2005/whisper-small-canto-lora

Whisper Small Canto - Chengyi Li

This model is a fine-tuned version of openai/whisper-small-lora on the Common Voice 24.0 - Cantonese dataset. It achieves the following results on the evaluation set:

• Best CER: 12.35

Model description

This model utilizes the LoRA fine-tuning technique.

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Done on an RTX 5060ti GPU

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Step	Epoch	Training Loss	Validation Loss	CER
1000	2.1552	0.0793	0.2843	13.3287
2000	4.3103	0.0192	0.3177	12.8122
3000	6.4655	0.0072	0.3567	12.7479
4000	8.6207	0.0016	0.3784	12.4832
5000	10.7759	0.0007	0.3973	12.4593
6000	12.9310	0.0004	0.4190	12.3509
7000	15.0862	0.0002	0.4235	12.5053
8000	17.2414	0.0001	0.4383	12.4446
9000	19.3966	0.0001	0.4425	12.5439
10000	21.5517	0.0001	0.4436	12.4354

Framework versions

• PEFT 0.18.1
• Transformers 4.52.0
• Pytorch 2.9.1+cu128
• Datasets 4.5.0
• Tokenizers 0.21.4