Hardware /

Apple M2 Ultra (76-core GPU)

14measured results
109best gen tok/s*
192 GBreported memory

*Headline excludes tiny/test models (under 0.6B params or TinyStories-style). Open in rig checker →

llama.cpp

ModelQuantGen tok/sPrompt tok/sSourceContext testedBuild
Llama 2 7BQ4_01091,412llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ4_01091,392llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ4_094.31,238llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ4_094.31,238llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ8_073.31,387llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ8_073.11,368llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ8_066.61,249llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BQ8_066.61,249llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BF1643.21,561llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BF1643.11,526llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BF1641.01,402llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Llama 2 7BF1641.01,402llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Qwen2.5 14B InstructQ4_K - Medium36.6471LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium36.5469LocalScore ↗4,352llamafile 0.9.2Details →

Ollama

No measured Ollama results yet. Submit yours →

MLX

No measured MLX results yet. Submit yours →

Includes results from LocalScore, a Mozilla Builders project by cjpais, mirrored in the cn0303/local-llm-speed-benchmarks dataset. LocalScore runs use llamafile, which is built on llama.cpp.