shoibo/whisper_IPA_final

whisper_IPA_mnx_lr2e5

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3020
• Wer: 49.8936
• Cer: 10.8822

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 200
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
9.8811	1.0	94	25.8793	1.7579	90.2119
4.1492	2.0	188	15.8005	1.1126	60.9485
2.7002	3.0	282	12.1085	0.9240	51.4632
2.0700	4.0	376	10.6037	0.8026	45.7114
1.7341	5.0	470	10.3937	0.7765	45.3078
1.3342	6.0	564	9.5888	0.8075	40.3633
1.0914	7.0	658	9.0639	0.7846	40.8678
0.9527	8.0	752	9.2913	0.7992	39.4551
0.7752	9.0	846	8.5914	0.7771	38.7487
2.2441	10.0	940	0.5908	77.4468	30.3030
1.3215	11.0	1034	0.3993	65.8511	16.7189
1.0320	12.0	1128	0.3606	61.1702	14.9425
0.8958	13.0	1222	0.3401	55.4255	12.7034
0.7670	14.0	1316	0.3020	49.8936	10.8822

Framework versions

• Transformers 5.15.0
• Pytorch 2.11.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2