Sep29-Mixat-whisper-lg-3-translation-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: 0.8625
• Wer: 46.4677

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
0.8851	0.4762	100	0.6177	57.9329
0.5912	0.9524	200	0.5425	55.1678
0.4614	1.4286	300	0.5272	49.1317
0.466	1.9048	400	0.5086	48.0863
0.3466	2.3810	500	0.5289	46.4003
0.3488	2.8571	600	0.5107	44.7311
0.2382	3.3333	700	0.5503	44.7648
0.2208	3.8095	800	0.5494	47.0578
0.1624	4.2857	900	0.5938	45.3718
0.1237	4.7619	1000	0.5893	45.4055
0.0966	5.2381	1100	0.6492	45.2032
0.0712	5.7143	1200	0.6321	43.5003
0.0614	6.1905	1300	0.6663	46.0968
0.0422	6.6667	1400	0.6621	45.1526
0.0423	7.1429	1500	0.6943	44.7142
0.0292	7.6190	1600	0.6971	45.5572
0.0311	8.0952	1700	0.7240	45.3212
0.022	8.5714	1800	0.7203	44.8828
0.0252	9.0476	1900	0.7415	46.6026
0.0186	9.5238	2000	0.7361	45.4224
0.0189	10.0	2100	0.7539	46.2148
0.0133	10.4762	2200	0.7797	44.9671
0.0188	10.9524	2300	0.7688	45.4392
0.0138	11.4286	2400	0.7763	44.7985
0.013	11.9048	2500	0.7762	45.0008
0.0121	12.3810	2600	0.7999	43.0787
0.0132	12.8571	2700	0.7931	43.7194
0.011	13.3333	2800	0.8111	46.0293
0.0113	13.8095	2900	0.7986	44.2084
0.0111	14.2857	3000	0.7936	43.0787
0.0097	14.7619	3100	0.8021	45.1357
0.0105	15.2381	3200	0.8137	46.2991
0.0101	15.7143	3300	0.8118	44.2590
0.0095	16.1905	3400	0.8126	43.8375
0.007	16.6667	3500	0.8326	45.1357
0.0077	17.1429	3600	0.8108	43.6520
0.0059	17.6190	3700	0.8436	44.6805
0.0071	18.0952	3800	0.8633	44.8997
0.0064	18.5714	3900	0.8487	44.2421
0.007	19.0476	4000	0.8321	45.0851
0.0057	19.5238	4100	0.8478	45.2875
0.0064	20.0	4200	0.8485	45.1189
0.0068	20.4762	4300	0.8531	44.7479
0.0073	20.9524	4400	0.8625	46.4677

Framework versions

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