Closed mbendjilali closed 11 months ago
Hi @mbendjilali, did you make sure you had no other process running on your GPU ? In particular, are you perhaps using a desktop computer's GPU ? If so, you do not want any graphical user interface taking space on it, so make sure you are running on a GUI-free session. You do need to have all 11G 100% available for the python process.
Second, if that is not enough and FRNN is the source of memory error, you could try one of the following:
datamodule.xy_tiling=5
(will subdivide large tiles into 5x5
smaller tiles)datamodule.knn=25
(number of neighbors for adjacency graph and geometric point features computation)datamodule.knn_r=10
(maximum radius for KNN search)Note that modifying these may produce different performance for the pretrained checkpoint you are using. So you may want to retrain your own model if you change parameters.
Hope that helps
Thank you very much for your tips, that worked perfectly !
Awesome ! Thanks for the feedback, I appreciate it 😊
Hello, I've been trying to run a training on dales using the command line
on a 2080Ti GPU with 11 Gb of RAM, but I always end up crashing because of a torch.cuda.OutOfMemoryError. I tried to tweak some of the parameters proposed in the README, but nothing does it. From my perspective, it looks like the training script always crash when calling frnn_grid_points. Here is an exemple traceback :