The 2l-4t.tar.xz
tar archive contains the weights of the μgat version with 2 layers and 4 tokens after 50 epochs of training on the arxiv dataset, without any additional finetuning on the tables dataset.
The 2l-4t.tar.xz
tar archive contains the weights of the μgat version with 2 layers and 4 tokens after 50 epochs of training on the arxiv dataset, without any additional finetuning on the tables dataset.