Same model, same Q4_K_M label: 5.02, 5.07 and 5.27 bits per weight
Summary
Picchio is a Python tool that measures local LLM performance, focusing on effective bits per weight, quantization, and the separation of prefill, decode, and wallclock timings. The article analyzes Qwen3.5-9B across multiple quantizers, reporting bits-per-weight values around 5.02 to 5.27 and discusses how to interpret lane measurements, silent CPU fallback, and OS GPU monitoring. It includes usage instructions, a command table, and notes on limits and licensing for local benchmarking.