wav2vec2-kashmiri-full-data

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: 0.9971
• Wer: 0.5078
• Cer: 0.1601

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 500
• num_epochs: 12
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.7859	200	3.4577	1.0	1.0
12.8852	1.5697	400	2.5259	1.0	0.7353
12.8852	2.3536	600	1.0384	0.8125	0.2773
2.4385	3.1375	800	0.8222	0.6533	0.2156
2.4385	3.9234	1000	0.7358	0.5958	0.1898
1.2812	4.7073	1200	0.7123	0.5749	0.1791
1.2812	5.4912	1400	0.7409	0.5552	0.1728
0.7982	6.2750	1600	0.7872	0.5438	0.1703
0.7982	7.0589	1800	0.8121	0.5399	0.1693
0.4679	7.8448	2000	0.8149	0.5356	0.1675
0.4679	8.6287	2200	0.8744	0.5244	0.1646
0.2934	9.4126	2400	0.9256	0.5176	0.1623
0.2934	10.1965	2600	0.9820	0.5158	0.1617
0.2009	10.9823	2800	0.9721	0.5127	0.1616
0.2009	11.7662	3000	0.9987	0.5076	0.1600

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1
• Datasets 3.2.0
• Tokenizers 0.21.0