ASCEND_Dataset_Model

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

• Loss: 3.9199
• Wer: 0.9540
• Cer: 0.9868

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: 8
• eval_batch_size: 1
• 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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
16.9063	1.0	687	4.7768	1.0	1.0
5.0252	2.0	1374	4.7004	1.0	1.0
4.9378	3.0	2061	4.6715	1.0	1.0
5.1468	4.0	2748	4.6605	1.0	1.0
4.9353	5.0	3435	4.6470	1.0	1.0
4.913	6.0	4122	4.6177	1.0	1.0
4.8034	7.0	4809	4.7699	1.0	1.0
4.6905	8.0	5496	4.3596	1.0	1.0
4.5251	9.0	6183	4.2670	1.0	1.0
4.4527	10.0	6870	4.2087	1.0	1.0
4.3731	11.0	7557	4.1950	0.9982	0.9997
4.3461	12.0	8244	4.2287	0.9928	0.9988
4.3224	13.0	8931	4.1565	0.9802	0.9971
4.2504	14.0	9618	4.1254	0.9619	0.9937
4.2196	15.0	10305	4.0377	0.9562	0.9913
4.1911	16.0	10992	4.0576	0.9601	0.9887
4.1079	17.0	11679	4.0630	0.9544	0.9857
4.1117	18.0	12366	4.0009	0.9558	0.9880
4.0324	19.0	13053	3.9245	0.9540	0.9877
3.9871	20.0	13740	3.9199	0.9540	0.9868

Framework versions

• Transformers 4.17.0
• Pytorch 1.10.0+cu111
• Datasets 2.0.0
• Tokenizers 0.11.6