René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Early deletion in `gw_ml.F90`","articleBody":"https://github.com/DataWaveProject/CAM/blob/8ef0f5a2a861c08eb61814514f3cc93dd383a3a5/src/physics/cam/gw_ml.F90#L123\n\nAren't we are deleting too early since the allocation of net_outputs is aliased by net_output_tensors? \n\nOr does the call to `torch_delete(net_output_tensors(1))` likely only destroy the PyTorch tensor wrapper, not the memory buffer it pointed to (which is net_outputs(:, i)). There is a check for zero outputs, so if this isn't satisfied wouldn't `net_outputs` be nonsense and therefore blow up? This doesn't tally with our observations (3 year runs were successful at the time). \n\nIf we moved to batching and pushed the deletions a little lower this wouldn't be a discussion anyway. \n\n\nMaybe @jatkinson1000 can help with this. ","author":{"url":"https://github.com/ma595","@type":"Person","name":"ma595"},"datePublished":"2025-11-11T10:46:33.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":3},"url":"https://github.com/37/CAM/issues/37"}