whisper-indian-lora

This model is a fine-tuned version of openai/whisper-medium on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0965
• Wer: 27.2815

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2753	0.9615	200	0.2702	22.9468
0.2379	1.9231	400	0.2266	25.8742
0.2321	2.8846	600	0.1956	27.9838
0.1932	3.8462	800	0.1700	22.5696
0.1724	4.8077	1000	0.1499	19.2853
0.1561	5.7692	1200	0.1316	32.9904
0.133	6.7308	1400	0.1191	25.8352
0.1117	7.6923	1600	0.1066	27.2093
0.102	8.6538	1800	0.1000	26.7368
0.1078	9.6154	2000	0.0965	27.2815

Framework versions

• PEFT 0.18.1
• Transformers 4.57.6
• Pytorch 2.10.0+cu128
• Datasets 4.8.3
• Tokenizers 0.22.2