OpenClaw + oMLX shows 0 cached tokens, but Hermes uses cache fine with the same local model, what am I missing?
Hey everyone, I’m trying to debug a weird prompt cache issue with OpenClaw + oMLX, and I’d appreciate help from anyone running local agents on MLX/oMLX. The short version is this: I’m running oMLX v0.3.8 on my Mac, serving: Qwen3.6-35B-A3B-RotorQuant-M…