arun100/whisper-base-hi-4

Whisper Base Hindi

This model is a fine-tuned version of arun100/whisper-base-hi-3 on the mozilla-foundation/common_voice_16_0 hi dataset. It achieves the following results on the evaluation set:

• Loss: 0.4681
• Wer: 27.6638

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
0.1251	13.16	1000	0.4681	27.6638
0.0812	26.32	2000	0.5046	28.2065
0.0584	39.47	3000	0.5393	28.3046
0.0441	52.63	4000	0.5639	28.4924
0.0392	65.79	5000	0.5734	28.5863

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0