speculative decoding in llama.cpp

[phasetrack]

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

It feels like the conversation has shifted hard toward inference and serving details lately, and honestly that makes sense once you start dealing with continuous batching.

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

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

[bytesupply76]

What exact setup are you running? Are you testing this on B200 or something else?

[thetagrid15]

What exact setup are you running? Are you testing this on B200 or something else?

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

[mellowwire98]

I'd approach it in a boring order. If you're on MLX, I'd confirm the version first and then reduce variables until the behavior is obvious. I keep promising myself this is the last rebuild.

[cozykite]

[quote]Interesting, I hadn't tried that. That matches what I saw on 7900 XTX. I got better results once I stopped chasing one metric and just looked a

Maybe, but I had the opposite result. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model. Could be a workload difference, but I wouldn't generalize too fast from one setup.

If you post your exact settings, people here can probably narrow it down pretty fast.

[dockridge]

Need a little more detail here. Is this local-only or are you routing anything externally?

[embershed99]

My first guess would be this: On 7900 XTX, the easiest win is often to back off the aggressive settings and see whether the problem survives. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[macroforge]

I'd approach it in a boring order. On 5090, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece.

If you post your exact settings, people here can probably narrow it down pretty fast.

[latentcraft]

I tested this tonight. I couldn't reproduce the slowdown until I pushed context much higher. I expected the model choice to matter more than it did.

[mangotinker73]

Ran a quick check on my box. On dual 3090 with Ollama, I saw basically the same behavior. I expected the model choice to matter more than it did.