simrun ❭ modular_reduced_model_inference ❭ strategy ❭ clear_memory

clear_memory¶

simrun.modular_reduced_model_inference.strategy.clear_memory()¶

Remove all weak references

Cupy frees GPU memory when all references are deleted As this is difficult to track, use the simrun.modular_reduced_model_inference.make_weakref() method, which storesall GPU arrays in the _WEAKREF_ARRAY_LIST and returns a weakref object. This can be used to interact with the data, but is not a reference. Therefore, it is sufficient to empty _WEAKREF_ARRAY_LIST, which frees the GPU memory. All weakref objects pointing to GPU arrays will then be invalidated.