S-Sethisak/Khmer_Orthography_Correction_Prahokbart

Khmer Orthographic Correction System using prohokbart

This model is a fine-tuned version of socheatasokhachan/khmerhomophonecorrector on the khmer-orthography-correction-dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.0935
• Cer: 0.0215
• Wer: 0.2767
• Bleu: {'score': 70.07249645533147, 'counts': [5795, 675, 216, 56], 'totals': [7574, 842, 301, 95], 'precisions': [76.51175072616847, 80.16627078384798, 71.76079734219269, 58.94736842105263], 'bp': 0.9818151194774639, 'sys_len': 7574, 'ref_len': 7713}

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: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer	Bleu
1.1071	1.0	842	0.3192	0.0882	0.6321	{'score': 40.58202238898504, 'counts': [3450, 441, 124, 24], 'totals': [7688, 956, 295, 76], 'precisions': [44.87513007284079, 46.12970711297071, 42.03389830508475, 31.57894736842105], 'bp': 0.9967534604238804, 'sys_len': 7688, 'ref_len': 7713}
0.3039	2.0	1684	0.1905	0.0666	0.4723	{'score': 54.74040402356386, 'counts': [4457, 535, 156, 33], 'totals': [7572, 840, 281, 71], 'precisions': [58.86159535129424, 63.69047619047619, 55.51601423487544, 46.478873239436616], 'bp': 0.9815510679220464, 'sys_len': 7572, 'ref_len': 7713}
0.1748	3.0	2526	0.1445	0.0374	0.3951	{'score': 57.62742438460407, 'counts': [4980, 581, 170, 37], 'totals': [7575, 843, 293, 82], 'precisions': [65.74257425742574, 68.92052194543298, 58.02047781569966, 45.1219512195122], 'bp': 0.9819471195901224, 'sys_len': 7575, 'ref_len': 7713}
0.1299	4.0	3368	0.1206	0.0315	0.3570	{'score': 62.991137552877646, 'counts': [5261, 618, 185, 43], 'totals': [7578, 846, 291, 82], 'precisions': [69.42465030351016, 73.04964539007092, 63.57388316151203, 52.4390243902439], 'bp': 0.9823430172934962, 'sys_len': 7578, 'ref_len': 7713}
0.0913	5.0	4210	0.1096	0.0270	0.3271	{'score': 63.321923252837784, 'counts': [5457, 626, 185, 40], 'totals': [7574, 842, 290, 79], 'precisions': [72.04911539477159, 74.34679334916865, 63.793103448275865, 50.63291139240506], 'bp': 0.9818151194774639, 'sys_len': 7574, 'ref_len': 7713}
0.0693	6.0	5052	0.1011	0.0243	0.3064	{'score': 68.35331355974593, 'counts': [5591, 641, 194, 45], 'totals': [7558, 826, 282, 75], 'precisions': [73.97459645408838, 77.60290556900726, 68.79432624113475, 60.0], 'bp': 0.9797007893576459, 'sys_len': 7558, 'ref_len': 7713}
0.0566	7.0	5894	0.0979	0.0233	0.2932	{'score': 66.9913256330908, 'counts': [5685, 652, 203, 49], 'totals': [7577, 845, 299, 89], 'precisions': [75.02969512999869, 77.15976331360947, 67.89297658862876, 55.056179775280896], 'bp': 0.9822110684960047, 'sys_len': 7577, 'ref_len': 7713}
0.0445	8.0	6736	0.0940	0.0226	0.2832	{'score': 66.04599096693936, 'counts': [5765, 679, 225, 64], 'totals': [7621, 889, 336, 124], 'precisions': [75.64624065083322, 76.37795275590551, 66.96428571428571, 51.61290322580645], 'bp': 0.9880006665646296, 'sys_len': 7621, 'ref_len': 7713}
0.0376	9.0	7578	0.0937	0.0223	0.2795	{'score': 66.16689942154542, 'counts': [5786, 684, 222, 62], 'totals': [7607, 875, 331, 122], 'precisions': [76.06152228210858, 78.17142857142858, 67.06948640483384, 50.81967213114754], 'bp': 0.9861621022557379, 'sys_len': 7607, 'ref_len': 7713}
0.0346	10.0	8420	0.0935	0.0215	0.2767	{'score': 70.07249645533147, 'counts': [5795, 675, 216, 56], 'totals': [7574, 842, 301, 95], 'precisions': [76.51175072616847, 80.16627078384798, 71.76079734219269, 58.94736842105263], 'bp': 0.9818151194774639, 'sys_len': 7574, 'ref_len': 7713}

Framework versions

• Transformers 4.57.2
• Pytorch 2.9.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1