wav2vec2-large-xls-r-300m-luo-googlefluers-5hr-v1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.7669
• Wer: 0.5508
• Cer: 0.1450

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: 0.0001
• 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: 40
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
14.2071	2.6667	100	5.4183	1.0	1.0
4.5199	5.3333	200	3.5672	1.0	1.0
3.2401	8.0	300	2.9414	1.0	1.0
2.912	10.6667	400	2.8496	1.0	1.0
2.293	13.3333	500	1.0939	0.8385	0.2482
0.7468	16.0	600	0.6503	0.601	0.1549
0.4431	18.6667	700	0.6416	0.5808	0.1534
0.2886	21.3333	800	0.6753	0.5793	0.1535
0.2085	24.0	900	0.6925	0.562	0.1467
0.1715	26.6667	1000	0.7211	0.5673	0.1477
0.1394	29.3333	1100	0.7347	0.5532	0.1430
0.1249	32.0	1200	0.7424	0.5543	0.1449
0.1131	34.6667	1300	0.7561	0.5588	0.1471
0.1034	37.3333	1400	0.7595	0.553	0.1445

Framework versions

• Transformers 4.42.3
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1