w2v-bert-bengali-model

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: 0.1759
• Wer: 0.1737

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: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• 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: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4772	0.4336	2000	0.3677	0.3733
0.3599	0.8671	4000	0.3126	0.3000
0.2908	1.3007	6000	0.2661	0.2659
0.2594	1.7342	8000	0.2433	0.2441
0.2041	2.1678	10000	0.2290	0.2279
0.1933	2.6013	12000	0.2107	0.2146
0.1676	3.0349	14000	0.2096	0.2117
0.1463	3.4685	16000	0.1956	0.1981
0.1332	3.9020	18000	0.1772	0.1848
0.106	4.3356	20000	0.1816	0.1788
0.1014	4.7691	22000	0.1759	0.1737

Framework versions

• Transformers 4.53.0
• Pytorch 2.7.0+cu126
• Datasets 2.18.0
• Tokenizers 0.21.1