pretrain-whisper-tiny-nepali-parliamentDS

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

• Cer: 300.2582
• Loss: 1.5477
• Wer: 323.4414

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• 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: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
2.3676	0.3641	300	282.7342	2.3985	175.9500
2.1685	0.7282	600	294.5234	2.1667	267.6756
1.9871	1.0922	900	109.7182	1.9758	115.7921
1.8147	1.4563	1200	126.2093	1.8518	164.4843
1.7212	1.8204	1500	99.7903	1.7706	115.6769
1.6459	2.1845	1800	83.0589	1.7044	104.2441
1.691	2.5485	2100	90.8084	1.6725	109.4917
1.5782	2.9126	2400	90.0721	1.6448	114.8544
1.5554	3.2767	2700	85.9736	1.6304	109.2614
1.5544	3.6408	3000	257.1743	1.6069	275.4236
1.4721	4.0049	3300	86.5591	1.5916	105.0995
1.4997	4.3689	3600	98.8776	1.5797	115.1012
1.4704	4.7330	3900	99.9795	1.5681	115.1834
1.5939	5.0971	4200	296.3592	1.6311	288.6988
1.6243	5.4612	4500	95.3902	1.6271	115.5124
1.5977	5.8252	4800	99.4989	1.6133	115.8085
1.5668	6.1893	5100	88.7682	1.6144	108.1099
1.5412	6.5534	5400	89.8599	1.5849	108.8830
1.4994	6.9175	5700	92.7772	1.5513	115.4137
1.426	7.2816	6000	83.7518	1.5310	100.2632
1.4677	7.6456	6300	300.2582	1.5477	323.4414

Framework versions

• Transformers 4.47.1
• Pytorch 2.6.0+xpu
• Datasets 3.2.0
• Tokenizers 0.21.0