Fine_Tunning_on_CV_Urdu_dataset

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_8_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.2389
• Wer: 0.7380

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
15.2352	1.69	100	4.0555	1.0
3.3873	3.39	200	3.2521	1.0
3.2387	5.08	300	3.2304	1.0
3.1983	6.78	400	3.1712	1.0
3.1224	8.47	500	3.0883	1.0
3.0782	10.17	600	3.0767	0.9996
3.0618	11.86	700	3.0280	1.0
2.9929	13.56	800	2.8994	1.0
2.785	15.25	900	2.4330	1.0
2.1276	16.95	1000	1.7795	0.9517
1.5544	18.64	1100	1.5101	0.8266
1.2651	20.34	1200	1.4037	0.7993
1.0816	22.03	1300	1.3101	0.7638
0.9817	23.73	1400	1.2855	0.7542
0.9019	25.42	1500	1.2737	0.7421
0.8688	27.12	1600	1.2457	0.7435
0.8293	28.81	1700	1.2389	0.7380

Framework versions

• Transformers 4.21.0
• Pytorch 1.11.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1