wav2vec2-large-xls-r-300m-Arabic-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0021
• Wer: 0.0191
• Cer: 0.0072

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: 0.0005
• train_batch_size: 16
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
17.5531	1.0	51	5.2205	1.0	1.0
3.9441	2.0	102	3.1378	1.0	1.0
3.1686	3.0	153	3.1191	1.0	1.0
3.1558	4.0	204	3.1101	1.0	1.0
3.1286	5.0	255	3.0171	1.0	1.0
3.0755	6.0	306	2.9542	1.0	1.0
2.9533	7.0	357	2.8221	1.0	1.0
2.5924	8.0	408	2.1453	1.0	0.9771
1.8657	9.0	459	1.1540	0.9094	0.7057
0.9519	10.0	510	0.4219	0.6767	0.2782
0.4752	11.0	561	0.1646	0.3416	0.0870
0.2402	12.0	612	0.0551	0.0899	0.0255
0.1512	13.0	663	0.0307	0.0586	0.0167
0.0906	14.0	714	0.0172	0.0541	0.0161
0.0711	15.0	765	0.0141	0.0444	0.0125
0.0561	16.0	816	0.0114	0.0269	0.0065
0.048	17.0	867	0.0090	0.0338	0.0110
0.0452	18.0	918	0.0072	0.0235	0.0080
0.0349	19.0	969	0.0073	0.0207	0.0062
0.0333	20.0	1020	0.0054	0.0183	0.0055
0.0275	21.0	1071	0.0050	0.0280	0.0087
0.0262	22.0	1122	0.0039	0.0251	0.0088
0.0241	23.0	1173	0.0039	0.0302	0.0110
0.0216	24.0	1224	0.0035	0.0243	0.0086
0.019	25.0	1275	0.0033	0.0250	0.0091
0.0178	26.0	1326	0.0027	0.0238	0.0089
0.0169	27.0	1377	0.0025	0.0220	0.0080
0.0168	28.0	1428	0.0024	0.0175	0.0060
0.015	29.0	1479	0.0021	0.0194	0.0071
0.0131	30.0	1530	0.0021	0.0191	0.0072

Framework versions

• Transformers 4.26.1
• Pytorch 1.13.1+cu116
• Datasets 2.10.0
• Tokenizers 0.13.2