Giters

martiansideofthemoon / style-transfer-paraphrase

Official code and data repository for our EMNLP 2020 long paper "Reformulating Unsupervised Style Transfer as Paraphrase Generation" (https://arxiv.org/abs/2010.05700).

Home Page:http://style.cs.umass.edu

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

RuntimeError: probability tensor contains either inf, nan or element < 0

HassanBinAli opened this issue · comments

commented

Hi! I have finetuned inverse paraphraser on custom dataset. Model has been saved. Now I am trying to load this model by running demo_paraphraser.py but it's giving
RuntimeError: probability tensor contains either inf, nan or element < 0
Essentially, this error is occurring when line number 313 of style_paraphrase/utils.py starts running. This line is
next_token = torch.multinomial(F.softmax(filtered_logits, dim=-1), num_samples=1)
Found some solutions but they didn't work. One of those solutions is to replace above line with the following
next_token = torch.multinomial(F.softmax(filtered_logits, dim=-1), num_samples=num_samples).
Also tried downgrading torch to 1.1.0 but the files in saved model with extension .bin are unable to be loaded with versions smaller than 1.6.0.
Any help in this regard will be appreciated.

commented

Hi! I have finetuned inverse paraphraser on custom dataset. Model has been saved. Now I am trying to load this model by running demo_paraphraser.py but it's giving RuntimeError: probability tensor contains either inf, nan or element < 0 Essentially, this error is occurring when line number 313 of style_paraphrase/utils.py starts running. This line is next_token = torch.multinomial(F.softmax(filtered_logits, dim=-1), num_samples=1) Found some solutions but they didn't work. One of those solutions is to replace above line with the following next_token = torch.multinomial(F.softmax(filtered_logits, dim=-1), num_samples=num_samples). Also tried downgrading torch to 1.1.0 but the files in saved model with extension .bin are unable to be loaded with versions smaller than 1.6.0. Any help in this regard will be appreciated.

I am also facing the same error. Here is the screenshot as well.
probability tensor inf error

commented

I don't think the model is training since it's outputting !!!!!!!! as the greedy sample.. Did you download the paraphrase generator correctly? Make sure you can paraphrase with the pretrained model first https://github.com/martiansideofthemoon/style-transfer-paraphrase/blob/master/README_terminal_demo.md