easycall-whisper-lg-3-Nov3

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0614
• Wer: 8.1563

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4391	0.0946	100	0.2666	33.1351
0.1608	0.1892	200	0.1262	17.0773
0.1116	0.2838	300	0.0969	12.4894
0.119	0.3784	400	0.1043	13.5939
0.0981	0.4730	500	0.0961	12.3619
0.0943	0.5676	600	0.0968	11.6822
0.087	0.6623	700	0.0886	11.5548
0.086	0.7569	800	0.0681	8.7935
0.0706	0.8515	900	0.0718	9.2608
0.067	0.9461	1000	0.0695	8.7935
0.0571	1.0407	1100	0.0682	8.3263
0.0596	1.1353	1200	0.0683	9.3883
0.0445	1.2299	1300	0.0677	8.0289
0.0577	1.3245	1400	0.0665	9.4308
0.0544	1.4191	1500	0.0633	8.5387
0.0491	1.5137	1600	0.0671	9.6856
0.0536	1.6083	1700	0.0617	8.6661
0.0406	1.7029	1800	0.0823	9.0059
0.0543	1.7975	1900	0.0614	8.1563

Framework versions

• Transformers 4.43.4
• Pytorch 2.4.1
• Datasets 3.0.0
• Tokenizers 0.19.1