Sep29-Mixat-whisper-lg-3-transliteration-0.1trainasval

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: 0.7875
• Wer: 39.7972

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
0.7619	0.4762	100	0.5335	54.2752
0.4908	0.9524	200	0.4641	49.9738
0.3846	1.4286	300	0.4498	43.4342
0.389	1.9048	400	0.4382	42.3151
0.2851	2.3810	500	0.4605	42.1927
0.2723	2.8571	600	0.4651	42.0878
0.202	3.3333	700	0.4855	40.9862
0.1731	3.8095	800	0.4809	41.3184
0.1243	4.2857	900	0.5475	40.6540
0.0988	4.7619	1000	0.5303	40.7064
0.0742	5.2381	1100	0.5775	40.6889
0.0531	5.7143	1200	0.5825	40.5316
0.0482	6.1905	1300	0.5976	41.3534
0.0368	6.6667	1400	0.6118	41.0911
0.0312	7.1429	1500	0.6439	42.0353
0.0242	7.6190	1600	0.6332	42.0528
0.0239	8.0952	1700	0.6684	39.3251
0.018	8.5714	1800	0.6527	42.3326
0.019	9.0476	1900	0.6736	40.7239
0.0153	9.5238	2000	0.6701	42.3326
0.0168	10.0	2100	0.7033	43.6790
0.0134	10.4762	2200	0.7028	40.2868
0.0141	10.9524	2300	0.6997	43.9063
0.0112	11.4286	2400	0.7055	42.1927
0.0118	11.9048	2500	0.7112	40.4266
0.0091	12.3810	2600	0.7509	41.5982
0.0106	12.8571	2700	0.7075	42.7872
0.0072	13.3333	2800	0.7263	43.3992
0.0096	13.8095	2900	0.7365	42.5249
0.0086	14.2857	3000	0.7722	42.0353
0.0099	14.7619	3100	0.7480	40.9862
0.0112	15.2381	3200	0.7422	40.9512
0.0076	15.7143	3300	0.7749	41.3709
0.0087	16.1905	3400	0.7505	39.9545
0.0073	16.6667	3500	0.7583	41.8780
0.0072	17.1429	3600	0.7541	41.0911
0.0063	17.6190	3700	0.7516	40.9337
0.0074	18.0952	3800	0.7798	41.2135
0.0067	18.5714	3900	0.7780	40.4791
0.0078	19.0476	4000	0.7596	41.2660
0.0061	19.5238	4100	0.7660	39.6048
0.0071	20.0	4200	0.7699	40.9862
0.0045	20.4762	4300	0.7855	41.4583
0.0055	20.9524	4400	0.7875	39.7972

Framework versions

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