Fixing Three Bugs to Make Qwen3.5-122B a Daily Driver on Mac Studio
Fixed three bugs that made Qwen3.5-122B a daily driver on Mac Studio
I spent three weeks debugging my local AI stack to run Qwen3.5-122B smoothly on my M3 Mac Studio. By switching from DS4 Flash and fixing critical cache leaks in my qMLX fork, I transformed a slow batch job into a responsive coding partner. The journey involved eliminating timestamp mismatches, handling interrupted replies, and cleaning up a poisoned checkpoint store to achieve sub-second context restoration.
That is not a chatbot, it is a batch job, and you go and make a cup of coffee while it thinks.
- hankbond
I appreciate the amount of detail in the post, I think it's a useful addition to the space.
That said, I have to read LLM output all day all the time, and I would implore you to take the time to explore your own voice a bit more.
> Two separate things then happened, and it is worth keeping them apart.
Is one of those phrases claude spits out nonstop.
- jval43
Impressive debugging skills, and thank you for the benchmarks. Now I'm wondering if mlx-engine / mlx-lm have these bugs too.
One minor thing: as you are concerned with honest numbers, the graphs should be logarithmic on the y-axis too (like they are on the x-axis). Otherwise it's hard to see whether the curve is sublinear or linear.
- adrianco
I pointed my Claude/Retort evaluator at this blog post and it ran an experiment to test what it found. We got better cache hit refill but it didn’t improve the overall coding results on my M5Pro/64GB system. Results in experiment-24 at GitHub.com/adrianco/retort