xlsr-big-kinnn

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Wer: 0.0510

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: 0.0004
• 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: 132
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.0626	2.1505	200	0.6872	0.5597
0.3624	4.3011	400	0.0468	0.1067
0.106	6.4516	600	0.0157	0.0641
0.0637	8.6022	800	0.0137	0.0625
0.0448	10.7527	1000	0.0086	0.0699
0.03	12.9032	1200	0.0027	0.0523
0.0264	15.0538	1400	0.0072	0.0559
0.0265	17.2043	1600	0.0042	0.0543
0.0266	19.3548	1800	0.0018	0.0553
0.0199	21.5054	2000	0.0083	0.0766
0.0139	23.6559	2200	0.0021	0.0569
0.0154	25.8065	2400	0.0033	0.0545
0.0145	27.9570	2600	0.0005	0.0510
0.0153	30.1075	2800	0.0017	0.0539
0.0144	32.2581	3000	0.0002	0.0527
0.0118	34.4086	3200	0.0024	0.0836
0.0118	36.5591	3400	0.0012	0.0575
0.0149	38.7097	3600	0.0046	0.0583
0.0125	40.8602	3800	0.0005	0.0571
0.0108	43.0108	4000	0.0003	0.0615
0.011	45.1613	4200	0.0010	0.0585
0.0085	47.3118	4400	0.0002	0.0510
0.0082	49.4624	4600	0.0003	0.0571
0.0076	51.6129	4800	0.0004	0.0607
0.0065	53.7634	5000	0.0001	0.0553
0.0058	55.9140	5200	0.0006	0.0512
0.0069	58.0645	5400	0.0020	0.0539
0.0046	60.2151	5600	0.0001	0.0537
0.0049	62.3656	5800	0.0001	0.0577
0.0038	64.5161	6000	0.0000	0.0615
0.0033	66.6667	6200	0.0000	0.0703
0.0048	68.8172	6400	0.0002	0.0711
0.0039	70.9677	6600	0.0000	0.0520
0.0048	73.1183	6800	0.0003	0.0523
0.0021	75.2688	7000	0.0001	0.0516
0.0024	77.4194	7200	0.0000	0.0510
0.0027	79.5699	7400	0.0000	0.0508
0.0014	81.7204	7600	0.0000	0.0504
0.0022	83.8710	7800	0.0000	0.0506
0.0019	86.0215	8000	0.0000	0.0520
0.0022	88.1720	8200	0.0000	0.0510
0.001	90.3226	8400	0.0000	0.0510
0.0012	92.4731	8600	0.0000	0.0510
0.0011	94.6237	8800	0.0000	0.0510
0.0014	96.7742	9000	0.0000	0.0510
0.0008	98.9247	9200	0.0000	0.0510

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1