paged attention observations with MLX

[labline]

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.

[cacheline]

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.

[vectorpoint]

Not sure I agree there. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.

[cacheline]

I'd start with the simpler stack.

[southtinker]

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.