search

vita-epfl / social-transmotion

[ICLR 2024] Official implementation of "Social-Transmotion: Promptable Human Trajectory Prediction" in PyTorch.
GNU Affero General Public License v3.0
34 stars 2 forks source link

LOSS is "NAN" #7

Closed 17771618951 closed 1 month ago

17771618951 commented 3 months ago

Hello, thanks for the author's works. I hava preprocessed the data, and this is my project dir: image I trained the JTA dataset using a single TeslaV100 GPU for 2 days, however, in the last 10~15 epoches, the loss turn to "NAN" . I wonder if the author faced this problem and how to sovle it? image

yanggao2000 commented 3 months ago

Hi, thanks for reaching out. The data directory looks okay. Regarding the loss issue, actually we haven't faced the issue before. Maybe could you double-check if you changed anywhere or followed all the provided instructions? Thanks.