arun100/whisper-base-bn

Whisper Base Bengali

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

• Loss: 0.2823
• Wer: 37.0102

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
0.4236	3.03	1000	0.4400	50.4684
0.3023	6.05	2000	0.3335	41.4718
0.2721	10.02	3000	0.3005	38.6209
0.2471	13.04	4000	0.2866	37.4478
0.2574	17.01	5000	0.2823	37.0102

Framework versions

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