Skip to content
llm-speed

llama3.2

4 workload results across 1 hardware configuration.

Fastest local config

103.1 decode tok/s

on M1 Pro (16-core GPU) + 16GB unified via ollama (Q8_0). see full run

Local runs (4 runs)

Runs from contributors' own machines via MLX, llama.cpp, vLLM, exllamav2, or ollama. Signed on the submitter's hardware.

M1 Pro (16-core GPU) + 16GB unifiedM1 Pro (16-core GPU) + 16GB unified

WorkloadBackendQuantdecode tok/sprefill tok/sTTFTRun
chat-shortollama@0.32.14Q8_0103.1tok/s8.77tok/s14,247msr_w-p7v--eync
chat-longollama@0.32.14Q8_081.35tok/s1,246.5tok/s2,529msr_w-p7v--eync
concurrent-decodeollama@0.32.14Q8_097.26tok/sno datano datar_w-p7v--eync
agent-traceollama@0.32.14Q8_093.14tok/s2,975.3tok/s516msr_w-p7v--eync

llama3.2 on hardware