I'm looking at /root/wandb-metadata.json for my runs. For the "cuda" field, the value is always null (File attached). Does this mean that the cudatoolkit is not found, and thus not optimizing the training for my GPU? watch nvidia-smi reports GPU uitilisation at ~100%. My conda environment was set up with the environment.yml provided, so the environment has
cudatoolkit 10.2.89 hfd86e86_1
Additionally, system wide, cuda/11.0 and cudnn/8.0.5-cuda11 are installed. I'm assuming backwards compatibility provided by cudnn8.0.5.
Hi,
Thanks for this incredible repo!
I'm looking at /root/wandb-metadata.json for my runs. For the "cuda" field, the value is always null (File attached). Does this mean that the cudatoolkit is not found, and thus not optimizing the training for my GPU?
watch nvidia-smi
reports GPU uitilisation at ~100%. My conda environment was set up with the environment.yml provided, so the environment has cudatoolkit 10.2.89 hfd86e86_1Additionally, system wide, cuda/11.0 and cudnn/8.0.5-cuda11 are installed. I'm assuming backwards compatibility provided by cudnn8.0.5.