10-13-2026, 04:29 PM
Thanks, that helps. I ended up in about the same place using LM Studio. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
2x3090 • llama.cpp • too much VRAM is still not enough