
Running Qwen3.8 27B as fast as possible on a MacBook Pro M5 Pro
From 20 to 60 tokens per second: benchmarking llama.cpp, Kapitsa and Splash with Qwen3.8 27B on a 48 GB MacBook Pro M5 Pro

From 20 to 60 tokens per second: benchmarking llama.cpp, Kapitsa and Splash with Qwen3.8 27B on a 48 GB MacBook Pro M5 Pro
I replayed 76 real requests from my self-hosted Hermes agents against Qwen3.8-Flash-Next, Qwen3.8-27B with SGLang + DFlash2 and Qwen3.6-35B-A3B with vLLM + MTP. Here are the numbers on TTFT, decode tok/s, concurrency and a blind quality review.

What happens when your smartwatch battery dies and the manufacturer won't sell replacement parts.

How I used Terraform and Ansible to build a disposable, production-like homelab on Alpine Linux
What started as a curiosity experiment became the best infrastructure decision I ever made. A tour of my selfhosted stack, the cloud services it replaced, and what I actually gained beyond the savings.

Automate repository dependency updates with Renovate on Forgejo