Tensor cores actually helping here

[emberrunner]

Thought I'd post a small recipe because a lot of threads ask similar things.

Just sharing a setup that ended up working reasonably well: FLUX, MLX, 4090, with a lot of attention on router policy.

Maybe useful as a starting point for newer folks. I keep promising myself this is the last rebuild.

[amberpatch25]

that fixed it for me too.

[phaseport36]

I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at KV cache pressure. Not glamorous advice, but repeatable baselines save a lot of time.

There are a few different conversations hiding inside this thread, so being explicit about the goal helps.

[orbitbench6]

Interesting, I hadn't tried that. The pattern around speculative decoding has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[llm_wren]

Interesting, I hadn't tried that. The pattern around speculative decoding has been pretty similar for me. The biggest change for me was simplifying th

Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was latency, not quality.

[amberpatch25]

I think the bottleneck is somewhere else. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model. Worth testing the simpler explanation before adopting the more interesting one.