torgo-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: 0.0877
• Wer: 5.6146

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
1.2844	0.3190	100	0.3944	28.6798
0.2992	0.6380	200	0.2182	17.4507
0.197	0.9569	300	0.1986	13.6571
0.1299	1.2759	400	0.1599	12.5190
0.1001	1.5949	500	0.1700	12.2155
0.1038	1.9139	600	0.1407	9.1806
0.0728	2.2329	700	0.1351	9.6358
0.0652	2.5518	800	0.1090	7.9666
0.0529	2.8708	900	0.1168	8.1942
0.0338	3.1898	1000	0.1132	6.8285
0.0358	3.5088	1100	0.0980	6.9044
0.0381	3.8278	1200	0.0820	6.7527
0.0245	4.1467	1300	0.0862	5.2352
0.0299	4.4657	1400	0.1068	5.9181
0.0261	4.7847	1500	0.0937	6.4492
0.0205	5.1037	1600	0.1019	7.2838
0.017	5.4226	1700	0.0990	5.6904
0.0115	5.7416	1800	0.0842	5.6146
0.018	6.0606	1900	0.1041	5.6904
0.0112	6.3796	2000	0.1135	7.1320
0.0174	6.6986	2100	0.0939	5.0835
0.0117	7.0175	2200	0.1092	5.9181
0.0121	7.3365	2300	0.0931	5.4628
0.0075	7.6555	2400	0.0974	5.6146
0.013	7.9745	2500	0.1142	5.7663
0.0063	8.2935	2600	0.1108	6.1457
0.0122	8.6124	2700	0.0929	5.6146
0.0125	8.9314	2800	0.0905	5.6146
0.0132	9.2504	2900	0.1202	6.1457
0.0132	9.5694	3000	0.0925	5.0835
0.0087	9.8884	3100	0.1028	5.0835
0.0043	10.2073	3200	0.0997	5.6146
0.004	10.5263	3300	0.0955	5.0835
0.007	10.8453	3400	0.0915	5.1593
0.0072	11.1643	3500	0.0814	5.3869
0.0069	11.4833	3600	0.0887	4.7800
0.0082	11.8022	3700	0.0940	5.6904
0.0105	12.1212	3800	0.0936	5.2352
0.0039	12.4402	3900	0.0966	5.3869
0.002	12.7592	4000	0.0820	4.7800
0.0032	13.0781	4100	0.0916	5.0835
0.0009	13.3971	4200	0.0877	5.6146

Framework versions

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