01a0683665f38d8e5e567b3b15ca98bf-Supplemental.pdf
–Neural Information Processing Systems
Whenm = 4, Viterbi inX4 returns "the opposite thing happened . In the main experiment table we showed latency/speedup results for WMT14 En-De. We used Adam optimizer [20], with betas 0.9 and 0.98.
Neural Information Processing Systems
Feb-7-2026, 07:27:18 GMT