distilbert-base-multilingual-cased-benali

This model is a fine-tuned version of distilbert/distilbert-base-multilingual-cased for Native Language Identification (NLI). In short, the task is to predict the native language (L1) of an author, given a text they wrote in their second language (L2). It achieves the following results on the evaluation set:

• Loss: 2.4213
• Accuracy: 0.3711

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

The model was trained on ~1M sentences in the following L2's: Czech, Chinese, English, Slovenian and Portuguese. The dataset covers more than 100 L1's.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• 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: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.4574	1.0	19659	2.4468	0.3545
2.3377	2.0	39318	2.3948	0.3683
2.2007	3.0	58977	2.3750	0.3720
2.0935	4.0	78636	2.3971	0.3719
1.9895	5.0	98295	2.4213	0.3711

Framework versions

• Transformers 4.54.1
• Pytorch 2.6.0+cu124
• Datasets 4.0.0
• Tokenizers 0.21.4