lucius-40/Whisper-bn-v2

whisper-Gandu

This model is a fine-tuned version of bangla-speech-processing/BanglaASR on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8103
• Wer: 20.3317
• Norm Levenshtein Similarity: 87.8463

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: 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_ratio: 0.1
• lr_scheduler_warmup_steps: 500
• num_epochs: 10
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.05

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Norm Levenshtein Similarity
1.5267	1.0	189	1.0251	47.5388	72.0330
0.9291	2.0	378	0.8935	35.9283	77.6067
0.8346	3.0	567	0.8532	28.8390	82.7667
0.7878	4.0	756	0.8247	23.9433	85.1927
0.7669	5.0	945	0.8160	24.5586	85.9831
0.7562	6.0	1134	0.8145	22.7394	85.7910
0.7504	7.0	1323	0.8084	24.5318	87.2333
0.7463	8.0	1512	0.8102	20.8400	88.0877
0.7441	9.0	1701	0.8101	19.7699	88.2547
0.7452	9.9496	1880	0.8103	20.3317	87.8463

Framework versions

• Transformers 4.48.3
• Pytorch 2.8.0+cu126
• Datasets 3.6.0
• Tokenizers 0.21.4