iqbalasrif/whisper-tiny-hyperparameter

hyperparameter

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

• Loss: 1.4506
• Wer: 0.6884
• Cer: 0.2050

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use 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: 100
• training_steps: 300

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.9694	0.1020	30	3.7782	1.8748	1.0887
3.3735	0.2041	60	2.9598	1.0019	0.4254
2.5449	0.3061	90	2.1989	0.8820	0.3221
1.9987	0.4082	120	1.8648	0.8004	0.2606
1.7671	0.5102	150	1.6909	0.7619	0.2312
1.6285	0.6122	180	1.5863	0.7336	0.2245
1.5475	0.7143	210	1.5251	0.7216	0.2213
1.4793	0.8163	240	1.4807	0.6942	0.2035
1.5013	0.9184	270	1.4582	0.6904	0.2057
1.4438	1.0204	300	1.4506	0.6884	0.2050

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1
• Datasets 3.0.1
• Tokenizers 0.20.0