OpenNMT Forum

Multi-Blue.perl(why Blue score is greater than one )

i have tried to visit papers about BLue score. it should be between 0 and one but my models evaluation gives BLEU = 2.46, 21.6/4.2/1.1/0.4 (BP=1.000, ratio=1.190, hyp_len=77812, ref_len=65403)

and other pair says
Use of uninitialized value in division (/) at OpenNMT-py/tools/multi-bleu.perl line 139, line 662.
BLEU = 0.00, 3.8/0.5/0.1/0.0 (BP=1.000, ratio=1.194, hyp_len=5685, ref_len=4760)

@tarekegn - It is multiplied by 100. In this sense, your first score is 0.0246, not good at all (unless human evaluation says otherwise).

okay bro I Think I must train some more steps…this is the result of the first 20 models.thank you I understood it now