Gabriellamin

Why is a2-split harder than a1-split?

2

Hi, According to the a2-split data and verify_split_tests.ipynb, I notice that 'red' or 'square' must appear in the command. 1, The paper claims a2-split aims to "test model performance on...

readme中提到的测试结果比论文中高出10个点，这是怎么回事，求解答 average reward | average turn | success rate -|-|- 31.0821 | 3.7336 | 0.9676