Table 7

Experimental results for the fine-tuned LongFormer model on the development set depending on the context window size and the label weight.

Window size
(nb tokens)
Label weight
(MASK, NO_MASK)
Rdi +qiERdiERqiPdi +qiWPdi +qi
32 (1,1) 0.875 0.987 0.850 0.612 0.643 
32 (5,1) 0.924 0.992 0.915 0.550 0.583 
32 (10,1) 0.954 0.994 0.952 0.470 0.507 
 
512 (1,1) 0.847 0.986 0.817 0.929 0.932 
512 (5,1) 0.914 0.993 0.906 0.858 0.866 
512 (10,1) 0.937 0.997 0.930 0.767 0.783 
 
4,096 (1,1) 0.860 0.988 0.847 0.925 0.929 
4,096 (5,1) 0.916 0.988 0.913 0.843 0.856 
4,096 (10,1) 0.935 0.993 0.936 0.795 0.811 
Window size
(nb tokens)
Label weight
(MASK, NO_MASK)
Rdi +qiERdiERqiPdi +qiWPdi +qi
32 (1,1) 0.875 0.987 0.850 0.612 0.643 
32 (5,1) 0.924 0.992 0.915 0.550 0.583 
32 (10,1) 0.954 0.994 0.952 0.470 0.507 
 
512 (1,1) 0.847 0.986 0.817 0.929 0.932 
512 (5,1) 0.914 0.993 0.906 0.858 0.866 
512 (10,1) 0.937 0.997 0.930 0.767 0.783 
 
4,096 (1,1) 0.860 0.988 0.847 0.925 0.929 
4,096 (5,1) 0.916 0.988 0.913 0.843 0.856 
4,096 (10,1) 0.935 0.993 0.936 0.795 0.811 
Close Modal

or Create an Account

Close Modal
Close Modal