indic-whisper-reverse-ml-mft-1-0-1-7

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0418
• Wer: 24.1165
• Cer: 4.3588

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: 64
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• 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: cosine
• lr_scheduler_warmup_steps: 116
• training_steps: 580

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.4154	0.9915	58	0.0491	25.3272	4.5829
0.1158	1.9744	116	0.0447	25.0	4.7898
0.0856	2.9573	174	0.0433	24.9673	4.6277
0.0846	3.9402	232	0.0428	24.7055	4.5726
0.0576	4.9231	290	0.0423	24.2801	4.3726
0.0579	5.9060	348	0.0420	24.2474	4.3760
0.0516	6.8889	406	0.0414	24.0183	4.3519
0.043	7.8718	464	0.0421	24.1492	4.4002
0.0454	8.8547	522	0.0420	23.9856	4.3691
0.0422	9.8376	580	0.0418	24.1165	4.3588

Framework versions

• Transformers 4.51.1
• Pytorch 2.7.1+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4