speculative decoding actually worth it with PyTorch

[cacheline]

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

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

Wondering if this is just where the whole space is heading now.

[cacheline]

Maybe, but I had the opposite result. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Could be a workload difference, but I wouldn't generalize too fast from one setup.

[cacheline]

Maybe, but I had the opposite result. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Could be a worklo

A couple questions before I guess too hard. What model, what runtime, and roughly what context size?

[quantyard]

That matches what I was seeing. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[cacheline]

Tried to reproduce it just now. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed.

[vrambandit]

Tried to reproduce it just now. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up b

same here on my side.

[cacheline]

Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it did. I keep promising myself this is the last rebuild.