I wanted to start a broader mistakes and lessons thread around GPUs for Local AI.
A useful angle is automation, but it also connects with future improvements. Different tools and workflows can produce very different results even when the goal is similar.
It would be interesting to compare...