kiranpantha/whisper-tiny-ne

Whisper Tiny Nepali - Kiran Pantha

This model is a fine-tuned version of openai/whisper-tiny on the OpenSLR54 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2933
• Wer: 53.7269
• Cer: 16.1186

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8115	0.3597	300	0.7467	92.9167	34.9897
0.4976	0.7194	600	0.4963	79.2130	26.2625
0.3874	1.0791	900	0.4198	71.5046	22.6696
0.3422	1.4388	1200	0.3797	67.5926	20.8896
0.3179	1.7986	1500	0.3467	63.9120	19.3959
0.2451	2.1583	1800	0.3299	62.1528	18.6950
0.2167	2.5180	2100	0.3224	60.6713	18.3977
0.2428	2.8777	2400	0.3085	59.6528	17.6196
0.1862	3.2374	2700	0.3057	57.6620	16.9113
0.1795	3.5971	3000	0.3007	57.5231	16.7792
0.1758	3.9568	3300	0.2935	55.8565	16.5297
0.1496	4.3165	3600	0.2960	55.8796	16.3792
0.156	4.6763	3900	0.2940	55.4398	16.4819
0.1235	5.0360	4200	0.2915	54.4444	16.0085
0.1311	5.3957	4500	0.2936	54.4676	16.2801
0.1136	5.7554	4800	0.2933	53.7269	16.1186

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cxx11.abi
• Datasets 3.2.0
• Tokenizers 0.20.3