distilbert-base-uncased trained for Semantic Textual Similarity in Catalan
This is a test model that was fine-tuned using the Catalan traduction of Spanish datasets from stsb_multi_mt in order to understand and benchmark STS models.
Model and training data description
This model was built taking distilbert-base-uncased and training it on a Semantic Textual Similarity task using a modified version of the training script for STS from Sentece Transformers (the modified script is included in the repo). It was trained using the Spanish datasets from stsb_multi_mt which are the STSBenchmark datasets automatically translated to other languages using deepl.com. and salt.gva.es. Refer to the dataset repository for more details.
Intended uses & limitations
This model was built just as a proof-of-concept on STS fine-tuning using Catalan data and no specific use other than getting a sense on how this training works.
How to use
You may use it as any other STS trained model to extract sentence embeddings. Check Sentence Transformers documentation.
Training procedure
Use the included script to train in Catalan the base model. You can also try to train another model passing it's reference as first argument. You can also train in some other language of those included in the training dataset.
Evaluation results
Evaluating distilbert-base-uncased on the Catalan test dataset before training results in:
Cosine-Similarity : Pearson: 0.3180 Spearman: 0.4014
While the fine-tuned version with the defaults of the training script and the Catalan training dataset results in:
Cosine-Similarity : Pearson: 0.7368 Spearman: 0.7288
Resources
- Training dataset stsb_multi_mt
- Sentence Transformers Semantic Textual Similarity
- Check sts_eval for a comparison with Tensorflow and Sentence-Transformers models
- Check the development environment to run the scripts and evaluation