ktrapeznikov/albert-xlarge-v2-squad-v2

Training on SQuAD V2 allows a model to score if a paragraph contains an answer . I did not sweep for best threshold . I used run_squad.py to fine-tuned the model on the dev set . The model was trained on 4 NVIDIA GeForce GeForce GTX 2080 Ti 11GBs . See huggingface.py for more details about the training and the training code . The training code is based on the model being used by the model of the model and the model is called Albert-xlarge-v2 . The code is described as 'albert' and 'quad' with the model's name or model's model . It is used to train,