speculative decoding in llama.cpp

[embershed99]

Posting this in case it saves someone else a little time.

I'm not trying to build a perfect architecture diagram. I just want a sane starting point around PyTorch and 7900 XTX.

I'm especially interested in mistakes worth avoiding.

[mangotinker73]

A couple questions before I guess too hard. Is this local-only or are you routing anything externally?

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

[vrambandit]

Appreciate the detail. 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.

[marbletrack]

I tested this tonight. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was latency, not quality.

[northcraft90]

Ran a quick check on my box. I couldn't reproduce the slowdown until I pushed context much higher. The surprise was that memory pressure showed up before throughput collapsed.

[vrambandit]

A couple questions before I guess too hard. Are you testing this on Threadripper box or something else?

[visionyard23]

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

[vrambandit]

following this because I'm in the same boat.

[papermixer4]

My first guess would be this: If you're on TensorRT-LLM, I'd confirm the version first and then reduce variables until the behavior is obvious.

[sanepeak48]

A couple things I'd check first. If you're on SGLang, I'd confirm the version first and then reduce variables until the behavior is obvious. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.