containment for model-driven automation: worth it?

[phasetrack]

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

Maybe useful as a starting point: PyTorch on GB200, keeping HBM constraints in mind from the start.

Feel free to tear it apart if you see something dumb.

For context, I'm not chasing leaderboard numbers. I mostly want the system to feel consistent day to day. At this point the real bottleneck is my patience.

[batchcraft]

Thanks, that helps. The pattern around prefill/decode split has been pretty similar for me. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing. At this point the real bottleneck is my patience.

[latentscope35]

That matches what I was seeing. The pattern around MoE serving has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[plainworks]

A couple questions before I guess too hard. Do you see the same thing with a smaller batch or shorter prompt?

[fastlab]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at structured output. Once you know the plain path works, then add the faster path back in piece by piece.

[thetanest]

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there.

[thetaarch43]

That matches what I was seeing. I ended up in about the same place using SGLang.

[quiettoken]

I'd approach it in a boring order. If you're on LM Studio, I'd confirm the version first and then reduce variables until the behavior is obvious. Once you know the plain path works, then add the faster path back in piece by piece.

[kernelmike]

Ran a quick check on my box. The issue went away once I changed one setting and restarted the stack cleanly. The surprise was that memory pressure showed up before throughput collapsed.

[southspark]

Tried to reproduce it just now. On B200 with PyTorch, I saw basically the same behavior.