arun100/whisper-small-bn-3

Whisper Base Bengali

This model is a fine-tuned version of arun100/whisper-base-hi-1 on the mozilla-foundation/common_voice_16_0 bn dataset. It achieves the following results on the evaluation set:

• Loss: 0.2754
• Wer: 36.2048

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-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.693	1.04	500	0.6937	69.6782
0.3979	3.03	1000	0.4168	48.0703
0.3429	5.01	1500	0.3527	42.8105
0.2907	6.05	2000	0.3225	40.4267
0.2761	8.03	2500	0.3039	38.8974
0.2637	10.02	3000	0.2921	37.7927
0.2507	12.0	3500	0.2846	37.0733
0.2397	13.04	4000	0.2793	36.6004
0.243	15.03	4500	0.2763	36.3503
0.2501	17.01	5000	0.2754	36.2048

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0