README.md

metadata

license: apache-2.0
tags:
  - generated_from_trainer
model-index:
  - name: urdumodel
    results: []

urdumodel

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: 0.4939
• Wer: 0.3698
• Cer: 0.1465

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: 24
• eval_batch_size: 24
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 96
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
2.8998	1.0	508	0.2832	1.0261	0.7004
0.7426	2.0	1016	0.2026	0.6532	0.5236
0.5694	3.0	1524	0.1799	0.5495	0.4611
0.4966	4.0	2032	0.1729	0.5361	0.4350
0.4555	5.0	2540	0.1684	0.5335	0.4266
0.4203	6.0	3048	0.1641	0.5040	0.4107
0.3951	7.0	3556	0.1579	0.5213	0.4037
0.3675	8.0	4064	0.1563	0.4949	0.3973
0.3555	9.0	4572	0.1581	0.4968	0.3978
0.3408	10.0	5080	0.1561	0.4827	0.3925
0.3286	11.0	5588	0.1524	0.5011	0.3858
0.3156	12.0	6096	0.1524	0.4871	0.3833
0.3047	13.0	6604	0.1499	0.4835	0.3774
0.2929	14.0	7112	0.1489	0.4844	0.3751
0.2912	15.0	7620	0.4929	0.3763	0.1486
0.2969	16.0	8128	0.4990	0.3749	0.1481
0.2946	17.0	8636	0.4943	0.3735	0.1485
0.2851	18.0	9144	0.4893	0.3717	0.1477
0.279	19.0	9652	0.4977	0.3693	0.1464
0.2718	20.0	10160	0.4939	0.3698	0.1465

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.0
• Datasets 2.4.0
• Tokenizers 0.12.1