whisper-small-santali-sanlish-frozen

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8507
• Wer: 45.2756
• Cer: 13.0534

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• 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: 240
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.6601	1.0	97	3.5435	88.8780	50.1312
1.9323	2.0	194	1.0238	63.3858	20.8749
0.7640	3.0	291	0.8547	56.5945	16.8854
0.5141	4.0	388	0.7974	49.4094	14.4882
0.3704	5.0	485	0.7753	48.1299	14.4882
0.2677	6.0	582	0.7642	46.5551	13.2633
0.1872	7.0	679	0.7691	52.9528	21.5398
0.1286	8.0	776	0.7709	47.3425	13.2283
0.0885	9.0	873	0.7775	45.4724	13.0184
0.0601	10.0	970	0.7860	52.8543	29.1514
0.0410	11.0	1067	0.7939	45.0787	12.6684
0.0292	12.0	1164	0.8127	45.3740	12.6509
0.0199	13.0	1261	0.8140	44.9803	13.0709
0.0148	14.0	1358	0.8224	44.4882	12.8784
0.0115	15.0	1455	0.8331	45.3740	13.0009
0.0098	16.0	1552	0.8363	56.2008	21.9598
0.0082	17.0	1649	0.8423	45.0787	13.0009
0.0072	18.0	1746	0.8474	56.1024	22.0472
0.0064	19.0	1843	0.8507	45.2756	13.0534

Framework versions

• Transformers 5.16.1
• Pytorch 2.11.0+cu128
• Datasets 4.0.0
• Tokenizers 0.23.1