whisper-tiny-khmer-v8-aug

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

• Loss: 0.2636
• Wer: 71.9453

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
1.0429	1.0	886	0.4805	98.9274
0.4846	2.0	1772	0.2949	103.0284
0.3615	3.0	2658	0.2385	83.4175
0.3033	4.0	3544	0.2234	88.8538
0.2717	5.0	4430	0.2046	78.0021
0.2435	6.0	5316	0.2006	77.0032
0.2202	7.0	6202	0.1941	82.3659
0.2054	8.0	7088	0.1884	77.1504
0.1911	9.0	7974	0.1871	74.7424
0.1788	10.0	8860	0.1875	75.4784
0.1685	11.0	9746	0.1900	75.8675
0.159	12.0	10632	0.1874	72.1241
0.1498	13.0	11518	0.1921	77.9390
0.1416	14.0	12404	0.1948	76.3407
0.134	15.0	13290	0.1941	73.7750
0.1261	16.0	14176	0.1992	74.0694
0.1187	17.0	15062	0.2045	75.2366
0.114	18.0	15948	0.2071	72.8496
0.1088	19.0	16834	0.2166	72.3975
0.1029	20.0	17720	0.2200	73.3754
0.098	21.0	18606	0.2238	72.3344
0.0944	22.0	19492	0.2252	73.0599
0.0897	23.0	20378	0.2344	72.9863
0.0866	24.0	21264	0.2387	73.0915
0.0834	25.0	22150	0.2442	73.8381
0.0803	26.0	23036	0.2478	74.9632
0.0769	27.0	23922	0.2522	74.3113
0.0739	28.0	24808	0.2582	74.0904
0.0723	29.0	25694	0.2618	72.4711
0.0702	29.9667	26550	0.2636	71.9453

Framework versions

• Transformers 4.47.0
• Pytorch 2.3.1
• Datasets 3.2.0
• Tokenizers 0.21.0