Open Servando1990 opened 4 years ago
In my case, I solved this issue (albeit in a different setting) by setting the block_size
to 512. If the block_size
is unspecified, it seems that the tokenizer's maximum size is used and since sometimes that can be really large -- the result is an empty dataset.
Hi @Shikib !
Do you mean block_size within the yml file?
Can't find this argument in the yml skeleton
Summary of the bug
Getting a Pytorch error num_samples should be a positive integer value but got num_samples=0. when using INRIA dataset to reproduce Fine-tuning notebook from the tutorial Solaris_FOSS4G_2019
Checking Pytorch forums and issues this is due to a wrong path to training data
https://github.com/thstkdgus35/EDSR-PyTorch/issues/185
PATH to training csv: training_data_csv: '/content/drive/My Drive/solaris/solaris_tutorials/Solaris_FOSS4G_2019/data/workshop_configs/inria_train.csv'
Screenshots
Training CSV structure
Environment information
solaris
version: 0.3.0