wav2vec2-large-xlsr-53-oria-Version2

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.7388
• Wer: 0.4380

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 70
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.1886	2.1505	100	5.8077	1.0
3.4942	4.3011	200	3.4903	1.0
3.4535	6.4516	300	3.4622	1.0
3.4533	8.6022	400	3.4484	1.0
3.5108	10.7527	500	3.4461	1.0
3.2951	12.9032	600	3.2130	1.0
1.1127	15.0538	700	1.0820	0.8127
0.6285	17.2043	800	0.7477	0.6510
0.5072	19.3548	900	0.6540	0.5758
0.3865	21.5054	1000	0.6212	0.5475
0.2889	23.6559	1100	0.6020	0.5204
0.2316	25.8065	1200	0.6334	0.5086
0.2105	27.9570	1300	0.6443	0.5007
0.2124	30.1075	1400	0.6318	0.4970
0.1916	32.2581	1500	0.6638	0.4918
0.1487	34.4086	1600	0.6562	0.4784
0.1742	36.5591	1700	0.6583	0.4808
0.1542	38.7097	1800	0.6618	0.4662
0.1015	40.8602	1900	0.6770	0.4652
0.1173	43.0108	2000	0.6982	0.4617
0.1162	45.1613	2100	0.7020	0.4575
0.1077	47.3118	2200	0.7189	0.4551
0.0914	49.4624	2300	0.7335	0.4536
0.1038	51.6129	2400	0.7180	0.4554
0.0814	53.7634	2500	0.7276	0.4524
0.0782	55.9140	2600	0.7411	0.4486
0.1001	58.0645	2700	0.7250	0.4474
0.0781	60.2151	2800	0.7226	0.4475
0.0677	62.3656	2900	0.7344	0.4426
0.0757	64.5161	3000	0.7341	0.4405
0.0636	66.6667	3100	0.7394	0.4390
0.0762	68.8172	3200	0.7388	0.4380

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1