langid-ner-xlm-v-base

This model is a fine-tuned version of facebook/xlm-v-base on the task 1 and task 2 of GUA-SPA@IberLEF 2023 shared task dataset. It achieves the following results on the evaluation set:

• Loss: 0.4369
• Precision: 0.8146
• Recall: 0.7939
• F1: 0.8041
• Accuracy: 0.9030

Model description

More information needed

Intended uses & limitations

• NER (PER, LOC, ORG)
• Token-based language identification (es, gn, mix, foreign)

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• 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
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	72	1.5829	0.5318	0.4706	0.4993	0.5808
No log	2.0	144	1.0009	0.5864	0.5495	0.5673	0.7504
No log	3.0	216	0.8142	0.6104	0.6124	0.6114	0.7966
No log	4.0	288	0.6806	0.6983	0.7047	0.7015	0.8404
No log	5.0	360	0.6066	0.7211	0.7223	0.7217	0.8448
No log	6.0	432	0.5607	0.7248	0.7357	0.7302	0.8672
0.9923	7.0	504	0.5318	0.7443	0.7450	0.7447	0.8802
0.9923	8.0	576	0.5072	0.7521	0.7584	0.7552	0.8806
0.9923	9.0	648	0.4955	0.7490	0.7584	0.7536	0.8796
0.9923	10.0	720	0.4924	0.7508	0.7609	0.7558	0.8802

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1