PyTorch compile got faster?

[cachebench]

This keeps coming up in conversation so I figured I'd start a thread.

One thing I keep noticing is that people underestimate structured output until they actually try to run this stuff for more than one afternoon.

Maybe obvious to some of you, but it clicked for me this week.

[northgarden37]

I'd start with the simpler stack.

[phasestream53]

I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at NVFP4. Not glamorous advice, but repeatable baselines save a lot of time.

[northgrid33]

What worked for me was pretty simple. If you're on LM Studio, I'd confirm the version first and then reduce variables until the behavior is obvious.

[deepseeked]

Thanks, that helps. The pattern around token cost has been pretty similar for me.

[thetarunner]

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at MoE serving. Not glamorous advice, but repeatable baselines save a lot of time.

[labpath]

Good point. That matches what I saw on GB200.

[routecraft]

Good point. That matches what I saw on GB200.

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

[labpath]

Did a small comparison this afternoon. The issue went away once I changed one setting and restarted the stack cleanly.

[rocmpath]

I'd approach it in a boring order. If you're on TensorRT-LLM, I'd confirm the version first and then reduce variables until the behavior is obvious. Not glamorous advice, but repeatable baselines save a lot of time.