@tommorris the results from local coding models are nearly as good, but very limited by how much VRAM you have and whether your hardware has optimisations for tensor compute. A good part of what you’re paying them is hardware rental. But if you have a M series Mac with at least 16G, or a 2020s discrete GPU with substantial VRAM, no need to rent, or mess with package managers. Just one of the OpenCode or Ollama desktop apps and your own machine.
undefined
Pete Wildsmith
@weargoggles@mastodon.social