BLEU component in RUBY, CodeBLEU

[vlomshakov]

Thanks for sharing metric implementations!
Usually BLEU is computed on a corpus, not individual sentences.
Maybe it's not problem at all for metric like RUBY or CodeBLEU.
But BLEU component will have another values - micro-average vs. macro-average.

Anyway I think you have already noticed this difference:

codegen-metrics/metrics_evaluation/metrics/codebleu/codebleu.py

Line 304 in 4b8ffd7

] # for some reason we need this exact amount of brackets for compute_bleu to work; I don't fully understand why and this might be related to the issue with the wrong BLEU computation