INFO: The model is being continuously updated.

The model is a multilingual-e5-base model fine-tuned with the task of semantic textual similarity in mind.

Model Training

The model has been fine-tuned on the German subsets of the following datasets:

• German paraphrase corpus by Philip May
• paws-x
• stsb_multi_mt

The training procedure can be divided into two stages:

• training on paraphrase datasets with the Multiple Negatives Ranking Loss
• training on semantic textual similarity datasets using the Cosine Similarity Loss

Results

The model achieves the following results:

• 0.920 on stsb's validation subset
• 0.904 on stsb's test subset