ericwoo_krBLUF On my 32 GB Mac mini (M2 Pro), going from an 8B dense model to a 14B dense model drops generation throughput from 26.1 tok/s to 14.7 tok/s — a real, felt slowdown. But two MoE models, qwen3-coder:30b and gpt-oss:20b, hit 40.0 tok/s and 32.8 tok/s respectively, both faster than the little 8B Llama … Read more
BLUF On my 32 GB M2 Pro Mac mini, Apple MLX models loaded noticeably faster and generated tokens faster than their llama.cpp equivalents in every test — for example, gemma4:26b hit 38.7 tok/s on MLX vs 30.1 tok/s on llama.cpp, and qwen3:27b hit 9.3 tok/s vs 7.7 tok/s. That said, the generation gains are real … Read more
BLUF On my 32 GB Mac mini (Apple M2 Pro), qwen3-coder:30b — a Mixture-of-Experts model with roughly 3B active parameters — hit 41 tok/s generation speed, compared to 14.2 tok/s for qwen2.5-coder:14b and 6.9 tok/s for the dense qwen2.5-coder:32b. The MoE architecture is doing something very different from dense 32B here, and the raw numbers … Read more
BLUF On my 32 GB Mac mini (M2 Pro), dense models scale predictably — going from 8B to 14B drops generation speed from 27.9 tok/s to 15.6 tok/s. But MoE architecture throws that rule out: qwen3-coder:30b hits 42.8 tok/s, faster than every other model I tested, including the 8B dense one. The Setup Hardware: Mac … Read more
BLUF On my M2 Pro Mac mini, the MLX backend consistently beats llama.cpp on both load time and generation speed — but the size of the gap depends heavily on the model. For Qwen3.6 27B (dense), MLX lands at 9.3 tok/s generation vs 7.7 tok/s for llama.cpp. For Gemma4 26B (MoE), the gap widens to … Read more
BLUF On my 32 GB Mac mini (Apple M2 Pro), qwen3-coder:30b — a 30B Mixture-of-Experts model with roughly 3B active parameters — hit 43.3 tok/s generation speed, well ahead of the dense qwen2.5-coder:32b at 6.6 tok/s and qwen2.5-coder:14b at 15.1 tok/s. If raw generation throughput is your priority on constrained hardware, the MoE architecture makes … Read more
The short version: On a Mac mini (Apple M2 Pro, 32 GB), the fastest model I tested was also the biggest — Qwen3-Coder 30B, a Mixture-of-Experts model with only ~3B active parameters, ran at 42.8 tokens/sec, beating a plain 8B model. Total parameter count barely predicts speed on this hardware; active parameters do. And the much-hyped … Read more
The Bill That Forces the Decision The moment that forces the AWS-vs-homelab decision usually isn’t a planning meeting — it’s a billing alert at the end of month three. A single t3.xlarge running continuously costs roughly $120/month in compute alone, but that’s the floor, not the ceiling. Stack a 100GB gp3 EBS volume (~$8), a … Read more
The Problem With Sending Your Traffic Data Somewhere Else GA4’s data sampling kicks in at surprisingly low traffic levels — and the threshold isn’t based on what would be statistically useful for you, it’s based on what’s computationally cheap for Google. A small e-commerce site running a conversion funnel report over a 90-day window will … Read more
Why Self-Hosted Operators Stop Trusting Google Analytics GA4’s event model is the first thing that breaks trust. The old Universal Analytics had one job: tell you how many people read a page. GA4 replaced that with a flexible event schema that’s genuinely powerful for e-commerce funnels — and genuinely annoying if you just want a … Read more