You may be lucky - people from NTT have published a model based on JParaCrawl - a large JP/EN parallel corpus, which can be used together with fairseq

http://www.kecl.ntt.co.jp/icl/lirg/jparacrawl/