JasmineJiali

Hi, nice repo and I have a little question when reading your code. Why did you take greedy decode results as baseline and use it to normalize random decode results?...