+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Evaluation & Testing (https://www.datasculpting.ai/inference/forumdisplay.php?fid=16)
+--- Forum: Evals (https://www.datasculpting.ai/inference/forumdisplay.php?fid=66)
+--- Thread: benchmarks that map to real use with vLLM (/showthread.php?tid=126)
Pages:
1
2
Re: benchmarks that map to real use with vLLM - thetanest - 09-06-2026
What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Mixtral and compare from there.
Re: benchmarks that map to real use with vLLM - batchsize - 09-09-2026
A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at speculative decoding. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
Re: benchmarks that map to real use with vLLM - quantghost - 09-10-2026
Interesting, I hadn't tried that. 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.
There are a few different conversations hiding inside this thread, so being explicit about the goal helps.
Re: benchmarks that map to real use with vLLM - northcraft90 - 09-10-2026
Tried to reproduce it just now. On Threadripper box with PyTorch, I saw basically the same behavior. The main thing that moved was latency, not quality.
Re: benchmarks that map to real use with vLLM - ggufwire22 - 09-10-2026
Quote:Tried to reproduce it just now. On Threadripper box with PyTorch, I saw basically the same behavior. The main thing that moved was latency, not qualit
following this because I'm in the same boat.
Re: benchmarks that map to real use with vLLM - sanevault - 09-10-2026
Maybe, but I had the opposite result. I get why that seems right, but on 4090 the bottleneck felt elsewhere. Worth testing the simpler explanation before adopting the more interesting one.