Akaash1/wav2vec-khmer-english-medium

training

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

• Loss: 0.6439
• Wer: 0.2868

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
3.5853	1.2139	500	3.5375	1.0
1.2399	2.4277	1000	0.9809	0.7112
0.873	3.6416	1500	0.6867	0.5484
0.6348	4.8554	2000	0.5878	0.4702
0.5296	6.0680	2500	0.5660	0.4503
0.3954	7.2819	3000	0.5330	0.3988
0.3495	8.4957	3500	0.5123	0.3782
0.3118	9.7096	4000	0.5323	0.3663
0.2783	10.9235	4500	0.5348	0.3542
0.2375	12.1361	5000	0.5576	0.3394
0.2023	13.3499	5500	0.5566	0.3415
0.2053	14.5638	6000	0.5710	0.3295
0.175	15.7776	6500	0.5684	0.3216
0.1643	16.9915	7000	0.5954	0.3080
0.1427	18.2041	7500	0.6128	0.3062
0.1432	19.4180	8000	0.6331	0.3075
0.1201	20.6318	8500	0.6385	0.3015
0.1101	21.8457	9000	0.6397	0.2911
0.1124	23.0583	9500	0.6459	0.2905
0.1123	24.2722	10000	0.6439	0.2868

Framework versions

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