Akaash1/wav2vec-khmer-english-high

training

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the ai (self-generated), openslr, S4E16_Podcast dataset. It achieves the following results on the evaluation set:

• Loss: 0.8121
• Wer: 0.3454

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: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.6927	0.8921	500	3.5857	1.0
1.7767	1.7832	1000	1.4485	0.8759
1.2204	2.6744	1500	1.0367	0.6912
1.0017	3.5656	2000	0.8688	0.5996
0.8473	4.4567	2500	0.8460	0.5507
0.9443	5.3479	3000	0.7532	0.5252
0.7366	6.2391	3500	0.7715	0.4859
0.6677	7.1302	4000	0.7055	0.4681
0.5689	8.0214	4500	0.6773	0.4518
0.4676	8.9135	5000	0.6610	0.4520
0.4313	9.8046	5500	0.7282	0.4293
0.3885	10.6958	6000	0.6898	0.4195
0.362	11.5870	6500	0.7006	0.4007
0.3515	12.4781	7000	0.7178	0.4036
0.3706	13.3693	7500	0.7403	0.3976
0.3226	14.2605	8000	0.7065	0.3956
0.2968	15.1517	8500	0.7231	0.3877
0.287	16.0428	9000	0.7221	0.3786
0.3053	16.9349	9500	0.7420	0.3665
0.2618	17.8260	10000	0.7300	0.3731
0.2438	18.7172	10500	0.7554	0.3656
0.2721	19.6084	11000	0.7574	0.3606
0.2279	20.4996	11500	0.7677	0.3587
0.2387	21.3907	12000	0.7974	0.3525
0.2458	22.2819	12500	0.8128	0.3490
0.219	23.1731	13000	0.8136	0.3517
0.2005	24.0642	13500	0.8053	0.3469
0.192	24.9563	14000	0.8121	0.3454

Framework versions

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