Open fabienmorgan opened 3 months ago
That's right. I got the same issue and made exactly the same modification to fix it. Thanks @fabienmorgan
I'm not a contributor to this repository so I'm not entirely sure if this solution hand-waves the actual problem, but it seemed to work for me too.
While trying to run Latent space optimization(LSO_eval.py) I got an error:
The error could be resolved in changing the datatype on Line 1306 from byte to bool. Current version
generated[-1].masked_fill_(unfinished_sents.byte(), self.eos_index)
My version
generated[-1].masked_fill_(unfinished_sents.bool(), self.eos_index)
The model was downloaded from the readme and the benchmarks were cloned with LFS from GitHub in the case of the PMLB benchmark and downloaded from the linked site in the case of the Feynman benchmark.
I hope that you can look into the bug and reply if my correction is correct and doesn't only resolve the symptoms.
Configurations: