speculative decoding actually worth it

[deltastation]

Did a bit of testing and thought I'd post the result here.

I tried this with DeepSeek using llama.cpp on 5090. The headline result was better than I expected, but MoE serving still ended up being the limiting factor.

Nothing scientific here, just practical notes from a real setup.

[cacheline]

subscribed, curious where this lands.

[emberrunner]

Did a small comparison this afternoon. On 5090 with PyTorch, I saw basically the same behavior. The main thing that moved was latency, not quality.