This model is a version of albert-base-v2, which is fine-tuned using MLM on ml-arxiv-papers dataset.