Akaash1/wav2vec-khmer-english-minor

training

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the ai(self-generated) dataset. It achieves the following results on the evaluation set:

• Loss: 1.2848
• Wer: 0.5239

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.8193	2.0134	300	3.6502	1.0
3.6873	4.0268	600	3.5860	1.0
3.6939	6.0403	900	3.5928	1.0
3.5887	8.0537	1200	3.5864	1.0
3.6768	10.0671	1500	3.5857	1.0
3.6779	12.0805	1800	3.5844	1.0
3.6155	14.0940	2100	3.5892	1.0
3.7153	16.1074	2400	3.5939	1.0
3.6172	18.1208	2700	3.5917	1.0
3.7787	20.1342	3000	3.6037	1.0
3.6783	22.1477	3300	3.5838	1.0
3.4817	24.1611	3600	3.3638	1.0
2.4234	26.1745	3900	2.1296	0.9889
2.1423	28.1879	4200	1.7221	0.9170
1.5984	30.2013	4500	1.5372	0.7765
1.3794	32.2148	4800	1.3687	0.7083
1.321	34.2282	5100	1.2705	0.6509
1.0296	36.2416	5400	1.2370	0.6185
0.9186	38.2550	5700	1.3053	0.5710
0.8659	40.2685	6000	1.3237	0.5583
0.8559	42.2819	6300	1.2585	0.5534
0.7367	44.2953	6600	1.2490	0.5396
0.7216	46.3087	6900	1.2701	0.5276
0.662	48.3221	7200	1.2848	0.5239

Framework versions

• Transformers 4.57.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.22.1