Llama-3.2-3B

7 records across 2 hardware.

hardwaremodelquantbackendtok/spp tok/sWttft sctx/tgdateclassflagssource
M4 Max (128 GB)Llama-3.2-3B Q8_0llama.cpp34.0–––40962026-10-17sourced–source
IQ-9075Llama-3.2-3B Q2_Kllama.cpp (Hexagon)27.33––––2026-10community–source
IQ-9075Llama-3.2-3B Q2_Kllama.cpp (Hexagon)26.18––––2026-10community–source
IQ-9075Llama-3.2-3B Q2_Kllama.cpp (Hexagon)24.92––––2026-10community–source
IQ-9075Llama-3.2-3B Q2_Kllama.cpp (Hexagon)24.47––––2026-10community–source
M4 Max (128 GB)Llama-3.2-3B Q8_0llama.cpp12.8–––327682026-10-17sourced–source
M4 Max (128 GB)Llama-3.2-3B Q8_0llama.cpp5.9–––1024002026-10-17sourced–source

← all models