m4 max mac studio

[drewwells]

Thanks for your blog post. I didn't know I could more than double token count by switching from ollama to omlx.

Metric	Ollama	oMLX
Total time	82.4s	35.5s
Gen tok/s	40.4	84.0
Prefill	14.6s	7.6s

oMLX is 132% faster (2.3x throughput) on sustained multi-turn workloads.

m4-max-128gb-40gpu_ollama_fa-kvq80_responses.md
m4-max-128gb-40gpu_ollama_fa-kvq80.md
m4-max-128gb-40gpu_omlx_fa-kvq80_responses.md
m4-max-128gb-40gpu_omlx_fa-kvq80.md

[arthware-dev]

Thats awesome! Thank you for your contributions. I try to add the numbers to the overview :-)