Inference · compare

Llama 3.2 3B vs MiniCPM5 1B.

Side by side only on machines of the same kind, so the difference is the model, not the computer. Each figure is the fastest build of that model on that machine.

2machine kinds both ran on
MachineLlama 3.2 3B tok/sMiniCPM5 1B tok/sContext A / B
L4 · 23 GB97Q4_K_M (4-bit) by unsloth208Q4_K_M (4-bit) by openbmb128K / 64K
GeForce RTX 3080 Ti · 12 GB99Q4_K_M (4-bit) by bartowski381Q4_K_M (4-bit) by openbmb64K / 4K