Sep26-Mixat-whisper-lg-3-transliteration

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: 0.7321
• Wer: 40.6571

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
0.7747	0.4292	100	0.4311	36.6994
0.4882	0.8584	200	0.4418	35.7241
0.3749	1.2876	300	0.4387	40.5617
0.3644	1.7167	400	0.4506	40.1608
0.3451	2.1459	500	0.4571	42.6225
0.2678	2.5751	600	0.4558	38.1490
0.2737	3.0043	700	0.4406	38.5621
0.1576	3.4335	800	0.4937	42.0456
0.1653	3.8627	900	0.4995	41.7987
0.1113	4.2918	1000	0.5667	41.4100
0.0957	4.7210	1100	0.5606	39.9237
0.0817	5.1502	1200	0.6160	41.6984
0.0534	5.5794	1300	0.6003	42.2313
0.0549	6.0086	1400	0.5908	40.9724
0.0315	6.4378	1500	0.6655	40.5031
0.0364	6.8670	1600	0.7179	43.4389
0.0278	7.2961	1700	0.6839	42.8009
0.0251	7.7253	1800	0.6803	42.9891
0.0228	8.1545	1900	0.7166	42.3047
0.0197	8.5837	2000	0.7321	40.6571

Framework versions

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