wav2vec2-urdu-finetuned-ASR

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

• Loss: 0.4310
• Wer: 0.3760

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
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• 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
1.0744	0.6969	50	0.5043	0.4740
0.8764	1.3902	100	0.4806	0.4469
0.7693	2.0836	150	0.4633	0.4067
0.6607	2.7805	200	0.4613	0.4072
0.5787	3.4739	250	0.4582	0.4004
0.5612	4.1672	300	0.4384	0.3804
0.5532	4.8641	350	0.4093	0.3730
0.5585	5.5575	400	0.4122	0.3872
0.527	6.2509	450	0.4309	0.3793
0.5722	6.9477	500	0.4310	0.3760

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu126
• Datasets 4.4.2
• Tokenizers 0.22.1