polyglot-tagger/multilabel-language-identification

Polyglot Tagger: Multi-label Language Identification

Refer to polyglot-tagger/language-identification. It is trained on the same dataset as a text-classifier rather than as a token classifier.

This model is a fine-tuned version of xlm-roberta-base. It achieves the following results on the evaluation set:

• Loss: 0.0123
• Precision: 0.9859
• Recall: 0.9831
• F1: 0.9845
• Accuracy: 0.9412

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 18
• total_train_batch_size: 576
• 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: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss	Precision	Recall
0.2186	0.2925	2500	0.8560	0.9651	0.0395	0.9778	0.9528
0.1331	0.5851	5000	0.0232	0.9803	0.9717	0.9760	0.9070
0.1044	0.8776	7500	0.0172	0.9828	0.9774	0.9801	0.9218
0.0851	1.1700	10000	0.0150	0.9844	0.9801	0.9822	0.9311
0.0783	1.4626	12500	0.0136	0.9859	0.9809	0.9834	0.9354
0.0705	1.7551	15000	0.0126	0.9861	0.9826	0.9843	0.9399
0.0692	2.0	17094	0.0123	0.9859	0.9831	0.9845	0.9412

Framework versions

• Transformers 5.5.4
• Pytorch 2.11.0+cu128
• Datasets 4.8.4
• Tokenizers 0.22.2