w2v-bert-Telugu-large

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2203
• Wer: 0.2210
• Cer: 0.0392

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: 5e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.9356	0.6579	300	0.5106	0.5659	0.1265
0.4276	1.3158	600	0.4152	0.4787	0.0953
0.3481	1.9737	900	0.3907	0.4076	0.0824
0.239	2.6316	1200	0.3014	0.3680	0.0660
0.1957	3.2895	1500	0.3159	0.3361	0.0629
0.1454	3.9474	1800	0.2517	0.2744	0.0489
0.1	4.6053	2100	0.2371	0.2621	0.0469
0.0748	5.2632	2400	0.2243	0.2469	0.0432
0.0453	5.9211	2700	0.2188	0.2381	0.0409
0.029	6.5789	3000	0.2203	0.2210	0.0392

Framework versions

• Transformers 4.43.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1