amitkangra2020/whisper-small-marathi

Whisper Small Marathi - MTech

This model is a fine-tuned version of openai/whisper-small on the MTech Marathi Audio (Custom) dataset. It achieves the following results on the evaluation set:

• Loss: 0.0191
• Wer: 3.8919

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: 4
• total_train_batch_size: 32
• 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: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6054	1.0	23	0.3230	50.4865
0.1552	2.0	46	0.0622	12.0
0.0369	3.0	69	0.0271	5.0811
0.0095	4.0	92	0.0217	4.1081
0.0047	5.0	115	0.0191	3.8919

Framework versions

• Transformers 4.53.3
• Pytorch 2.6.0+cu124
• Datasets 4.4.1
• Tokenizers 0.21.2