wav2vec2bert-bengali-clean-audio-1e5

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: 0.9749
• Wer: 0.4271
• Cer: 0.1097

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: 1e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
11.3054	1.2675	100	5.1693	1.0430	0.6356
3.2001	2.5350	200	1.6331	0.6407	0.1828
2.4474	3.8025	300	1.2845	0.5187	0.1418
1.8174	5.0637	400	1.1087	0.4875	0.1327
1.4476	6.3312	500	1.0288	0.4508	0.1175
1.3777	7.5987	600	0.9930	0.4278	0.1123
1.1010	8.8662	700	0.9749	0.4271	0.1097

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.8.3
• Tokenizers 0.22.2