gec-score-model

This model is a fine-tuned version of intfloat/multilingual-e5-base on the peterua/OmniGEC-ModelTraining dataset.

Training script is available here: https://github.com/lapa-llm/lapa-llm/blob/main/pretraining/quality-classifiers/gec_score.py

It achieves the following results on the evaluation set:

• Loss: 0.1941
• Precision: 0.7031
• Recall: 0.7030
• F1 Macro: 0.7030
• Accuracy: 0.7030

Model description

This model outputs a score how grammatical correct is the provided text.

Intended uses & limitations

Pretraining data filtering.

Training and evaluation data

Training script is located here: https://github.com/lapa-llm/lapa-llm/blob/main/pretraining/quality-classifiers/gec_score.py

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 32
• eval_batch_size: 128
• seed: 0
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 256
• total_eval_batch_size: 1024
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 60

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1 Macro	Accuracy
No log	0	0	0.2963	0.5503	0.5013	0.3409	0.5013
0.2297	7.4074	400	0.2287	0.6545	0.6336	0.6208	0.6336
0.2051	14.8148	800	0.2041	0.6722	0.6660	0.6630	0.6660
0.1957	22.2222	1200	0.1982	0.6889	0.6885	0.6883	0.6885
0.1939	29.6296	1600	0.1963	0.6971	0.6964	0.6962	0.6964
0.1916	37.0370	2000	0.1946	0.7005	0.7004	0.7004	0.7004
0.1907	44.4444	2400	0.1944	0.7018	0.7017	0.7017	0.7017
0.1888	51.8519	2800	0.1944	0.6990	0.6984	0.6982	0.6984
0.1884	59.2593	3200	0.1941	0.7031	0.7030	0.7030	0.7030

Framework versions

• Transformers 4.56.1
• Pytorch 2.6.0a0+ecf3bae40a.nv25.01
• Datasets 4.0.0
• Tokenizers 0.22.0