It is technically possible to operate BerkeleyDb in-memory, but this option is not available in the current wrapper classes. Uncompressed models tend to be huge; however, larger machines could benefit from reading vectors from memory directly instead of accessing the disk. The caching option serves this purpose to some extent, but consumes memory very fast.
It is technically possible to operate BerkeleyDb in-memory, but this option is not available in the current wrapper classes. Uncompressed models tend to be huge; however, larger machines could benefit from reading vectors from memory directly instead of accessing the disk. The caching option serves this purpose to some extent, but consumes memory very fast.