Dzmitry Bahdanau
Dzmitry Bahdanau
I really appreciate that you responded so quickly, @ruiqi-zhong. I am also happy hear that you want to release a better version of the dataset. My team and I are...
I think I have found another issue. While I can see that a lot of effort was put into rewriting the original SQL queries from text2sql-data into Spider SQL, there...
Hi Ruiqi, thanks a lot for your continued effort to make this test-suite work correctly for classical datasets. Unfortunately, I think I have found another issue, this time probably the...
It is the case 3 that happens, namely: > The classical.pkl is missing certain combinations of natural language + literals from Jonathans' repo. I would argue this is a significant...
Great, thank you so much! I am looking forward to the new release.
Ok, thanks for confirming!
Same question here. According to `TRAINING.md`, in Step 2 "we train the execution engine, using programs predicted from the program generator in the previous step". In `train_model.py` line 238 says...
Yes, it's the learning rate. It should be decreased to 1e-5, and then step 2 works. Note, that it will indeed use the groundtruth programs in Step 2.
I get smth like 95-96%, which is what is reported in the paper.
Can you try training longer?