lepton runs with default settings for the platform
lepton-avx is compiled avx2 extensions and requires a compatible machine to run. It should be maginally faster on newer machines (eg if they don't downclock when using extended width vectors)
lepton-scalar avoids vector instructions altogether (especially platforms like javascript that don't have vector instructions)
lepton-slow-best-ratio defaults to single threaded behavior, allowing learned-buckets to be learned over the whole image rather than just a slice of it
lepton runs with default settings for the platform lepton-avx is compiled avx2 extensions and requires a compatible machine to run. It should be maginally faster on newer machines (eg if they don't downclock when using extended width vectors) lepton-scalar avoids vector instructions altogether (especially platforms like javascript that don't have vector instructions) lepton-slow-best-ratio defaults to single threaded behavior, allowing learned-buckets to be learned over the whole image rather than just a slice of it