This model has been trained using 3M by we have not seen an progress on the validation Cosine Similarity, so we have increased the lr to 3-e5
This model has been trained using 3M by we have not seen an progress on the validation Cosine Similarity, so we have increased the lr to 3-e5