+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Optimization & Performance (https://www.datasculpting.ai/inference/forumdisplay.php?fid=10)
+--- Forum: Context / KV Cache (https://www.datasculpting.ai/inference/forumdisplay.php?fid=41)
+--- Thread: paged attention observations with MLX (/showthread.php?tid=753)
Pages:
1
2
Re: paged attention observations with MLX - labline - 10-11-2026
That matches what I was seeing. I don't think your conclusion is far off. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
Re: paged attention observations with MLX - cacheline - 10-11-2026
Quote:That matches what I was seeing. I don't think your conclusion is far off. I got better results once I stopped chasing one metric and just looked at wh
That matches what I was seeing. That matches what I saw on 5090. The biggest change for me was simplifying the stack instead of adding one more optimization layer.
Re: paged attention observations with MLX - vectorpoint - 10-11-2026
Not sure I agree there. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.
Re: paged attention observations with MLX - cacheline - 10-11-2026
I'd start with the simpler stack.
Re: paged attention observations with MLX - southtinker - 10-12-2026
Quote:I'd start with the simpler stack.
That matches what I was seeing. I don't think your conclusion is far off. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.