|
--- |
|
license: mit |
|
base_model: xlm-roberta-base |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: xlm-roberta-base-finetuned-panx-all |
|
results: [] |
|
--- |
|
|
|
# xlm-roberta-base-finetuned-panx-all |
|
|
|
This model is a fine-tuned version of [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the XTREME PANX dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.1758 |
|
- F1 Score: 0.8558 |
|
|
|
## Model description |
|
|
|
This model is a fine-tuned version of xlm-roberta-base on a concatenated dataset combining multiple languages, specifically German (de) and French (fr). The model has been trained for token classification tasks and achieves competitive F1-scores across various languages. |
|
|
|
## Intended uses |
|
|
|
Named Entity Recognition (NER) tasks across multiple languages. |
|
Token classification tasks that benefit from multilingual training data. |
|
|
|
## Limitations |
|
|
|
Performance may vary on languages not seen during training. |
|
The model is fine-tuned on specific datasets and may require further fine-tuning or adjustments for other tasks or domains. |
|
|
|
## Training and evaluation data |
|
|
|
The model was fine-tuned on a combination of German and French datasets, with the training data shuffled and concatenated to form a multilingual corpus. Additionally, the model was evaluated on multiple languages, showing robust performance across different linguistic datasets. |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-05 |
|
- train_batch_size: 24 |
|
- eval_batch_size: 24 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 3 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | F1 Score | |
|
|:-------------:|:-----:|:----:|:---------------:|:--------:| |
|
| 0.299 | 1.0 | 835 | 0.2074 | 0.8078 | |
|
| 0.1587 | 2.0 | 1670 | 0.1705 | 0.8461 | |
|
| 0.1012 | 3.0 | 2505 | 0.1758 | 0.8558 | |
|
|
|
### Evaluation results |
|
|
|
The model was evaluated on multiple languages, achieving the following F1-scores: |
|
|
|
| Evaluated on | de | fr | it | en | |
|
|:-------------:|:-----:|:----:|:---------------:|:--------:| |
|
| Fine-tune on | | | | | |
|
| de |0.8658 | 0.7021 | 0.6877 | 0.5830 | |
|
| each |0.8658 | 0.8411 | 0.8180 | 0.6870 | |
|
| all |0.8685 | 0.8654 | 0.8669 | 0.7678 | |
|
|
|
### Framework versions |
|
|
|
- Transformers 4.41.1 |
|
- Pytorch 2.3.0+cu121 |
|
- Datasets 2.19.1 |
|
- Tokenizers 0.19.1 |
|
|
