whisper-tiny-khmer-v7

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

• Loss: 0.4052
• Wer: 101.1056

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9688	1.0	984	0.5047	102.8469
0.4468	2.0	1968	0.3500	100.6910
0.342	3.0	2952	0.2941	102.3494
0.2872	4.0	3936	0.2746	106.2742
0.2558	5.0	4920	0.2712	106.6888
0.2333	6.0	5904	0.2632	100.0553
0.212	7.0	6888	0.2624	103.1233
0.1958	8.0	7872	0.2677	99.8894
0.1848	9.0	8856	0.2733	99.2537
0.1733	10.0	9840	0.2767	111.4704
0.1632	11.0	10824	0.2843	104.5053
0.1543	12.0	11808	0.2837	100.6357
0.1448	13.0	12792	0.2883	98.9221
0.1364	14.0	13776	0.2939	100.4975
0.1289	15.0	14760	0.3024	100.1935
0.1218	16.0	15744	0.3071	101.8242
0.1167	17.0	16728	0.3153	101.4925
0.1115	18.0	17712	0.3211	99.8894
0.1053	19.0	18696	0.3365	100.3040
0.1003	20.0	19680	0.3418	100.6910
0.0952	21.0	20664	0.3502	102.0453
0.0924	22.0	21648	0.3556	102.2388
0.0882	23.0	22632	0.3611	99.7789
0.0848	24.0	23616	0.3714	100.0829
0.0809	25.0	24600	0.3705	101.5202
0.0791	26.0	25584	0.3761	99.5854
0.0761	27.0	26568	0.3834	100.3040
0.0733	28.0	27552	0.3965	102.1282
0.0708	29.0	28536	0.4027	101.2161
0.0689	30.0	29520	0.4052	101.1056

Framework versions

• Transformers 4.47.0
• Pytorch 2.3.1
• Datasets 3.1.0
• Tokenizers 0.21.0