Couldn't reproduce the multiWOZ 2.1 results #47
Description
Hello, I had run the baseline code you've provided. The joint goal accuracy goes up to only 0.12 at 50000-ckpt. when I trained further to 100000 or even to 200000, the joint goal accuracy dropped to near zero. I evaluate the predicted state using your evaluation script. Can you give me some hints about how you trained the model (like which global step can reproduce the result? )?
I followed the instructions in the README.md directly.