List your company

Machine Translation on Japanese-English (test)

27.2BLEUVanilla Decoding
-0.8631526.42247413.708120.993726Jan 23, 2017Mar 24, 2018May 23, 2019Jul 22, 2020Sep 20, 2021Nov 19, 2022Jan 19, 2024

Evaluation results

MethodLinks
Vanilla DecodingEvaluation Protocol=3-shot, Backbone=Mistral 7B, Tokens/sec=36.8, Speed Up=1.00x2024.0127.2
MuMoEvaluation Protocol=3-shot, Backbone=Mistral 7B, Tokens/sec=66.3, Speed Up=1.80x2024.0126.9
1.5-entmaxactivation=1.5-entmax2019.0822.13
alpha-entmaxactivation=alpha-entmax2019.0821.74
softmaxactivation=softmax2019.0821.57
MoE-MultiParameters=8.7B, ops/timestep=102M, training time=12 days, hardware=64 k40s2017.010.2591
GNMT-MonoParameters=278M/model, ops/timestep=212M, training time=various2017.010.2341
GNMT-MultiParameters=278M, ops/timestep=212M, training time=21 days, hardware=96 k20s2017.010.2162