iqbalasrif/whisper-tiny-reset-parameter

from-scratch

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: 10.8537
• Wer: 6.4322
• Cer: 9.6337

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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: 30
• training_steps: 150

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
10.8562	0.2041	30	10.8560	17.0625	24.9361
10.8557	0.4082	60	10.8553	20.3833	28.0200
10.8549	0.6122	90	10.8545	11.6823	17.2418
10.8542	0.8163	120	10.8539	10.5108	15.8910
10.8538	1.0204	150	10.8537	6.4322	9.6337

Framework versions

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