w2v-bert-2.0-khmer-v4

This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3175
• Wer: 0.2116
• Cer: 0.0573

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.4211	0.9897	72	3.4997	1.0	0.9974
2.3724	1.9931	145	1.1296	0.7829	0.2773
0.8154	2.9966	218	0.6209	0.5044	0.1551
0.497	4.0	291	0.4820	0.4117	0.1175
0.3677	4.9897	363	0.4351	0.3663	0.1038
0.2837	5.9931	436	0.4207	0.3516	0.1000
0.2396	6.9966	509	0.4339	0.3431	0.0953
0.1857	8.0	582	0.3432	0.2784	0.0750
0.1083	8.9897	654	0.3444	0.2422	0.0659
0.0554	9.8969	720	0.3175	0.2116	0.0573

Framework versions

• Transformers 4.46.3
• Pytorch 2.11.0+cu128
• Datasets 3.1.0
• Tokenizers 0.20.3