search

OFA-Sys / gsm8k-ScRel

Codes and Data for Scaling Relationship on Learning Mathematical Reasoning with Large Language Models
https://arxiv.org/abs/2308.01825
215 stars 16 forks source link

test.py 里面use_diverse_beam 和do_sample 都默认是False? #5

Closed bestpredicts closed 1 year ago

bestpredicts commented 1 year ago

这两个值都是false的情况下，模型每次生成的结果应该是固定的？

GanjinZero commented 1 year ago

greedy sample eval性能最好