M3 Ultra 20+8

3 records across 1 models (quants: F16, Q4_0, Q8_0).

hardwaremodelquantbackendtok/spp tok/sWttft sctx/tgdateclassflagssource
M3 Ultra 20+8llama 7B mostly Q4_0llama.cpp (Metal)88.41073.09––tg1282025-03-14community–source
M3 Ultra 20+8llama 7B mostly Q8_0llama.cpp (Metal)63.551085.76––tg1282025-03-14community–source
M3 Ultra 20+8llama 7B mostly F16llama.cpp (Metal)42.241121.8––tg1282025-03-14community–source

All rows cite their source. Provenance: community 40018 sourced 1085

← all hardware