10-19-2026, 12:55 AM
Interesting, I hadn't tried that. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.
2x3090 • llama.cpp • too much VRAM is still not enough
