Skip to content

High VRAM usage when predicting large number of structures #18

Description

@Alexsp32

MACEModels has quite a high memory usage after doing batch inference for a large number of structures. It feels like there must be options to garbage collect sooner, especially after inference has finished since data must have been transferred back to the host device by then.

No priority since all current projects are stuck on v1.1.1

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions