nepali-gec-error-type-classifier

This model is a fine-tuned version of IRIIS-RESEARCH/RoBERTa_Nepali_125M on the arrow dataset. It achieves the following results on the evaluation set:

• Loss: 0.1340
• Accuracy: 0.9502
• Macro F1: 0.6346
• Num Error Tokens: 1934543

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: 2e-06
• train_batch_size: 1024
• eval_batch_size: 1024
• seed: 42
• 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
• lr_scheduler_warmup_steps: 500
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Macro F1	Num Error Tokens
0.5944	0.1574	1000	0.4266	0.8294	0.5394	1934543
0.3682	0.3149	2000	0.2821	0.8919	0.5882	1934543
0.2972	0.4723	3000	0.2351	0.9111	0.6027	1934543
0.2618	0.6297	4000	0.2077	0.9221	0.6109	1934543
0.2379	0.7872	5000	0.1900	0.9292	0.6163	1934543
0.2217	0.9446	6000	0.1780	0.9339	0.6198	1934543
0.2066	1.1020	7000	0.1675	0.9378	0.6228	1934543
0.1977	1.2594	8000	0.1604	0.9405	0.6248	1934543
0.192	1.4169	9000	0.1546	0.9425	0.6263	1934543
0.1836	1.5743	10000	0.1502	0.9443	0.6276	1934543
0.18	1.7317	11000	0.1466	0.9456	0.6286	1934543
0.1763	1.8892	12000	0.1431	0.9469	0.6296	1934543
0.172	2.0466	13000	0.1408	0.9476	0.6301	1934543
0.1672	2.2040	14000	0.1383	0.9486	0.6317	1934543
0.1673	2.3615	15000	0.1368	0.9492	0.6321	1934543
0.1646	2.5189	16000	0.1355	0.9496	0.6333	1934543
0.1629	2.6763	17000	0.1349	0.9499	0.6327	1934543
0.1631	2.8338	18000	0.1342	0.9501	0.6337	1934543
0.1621	2.9912	19000	0.1340	0.9502	0.6346	1934543

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1