@hopeless the quality of the models cannot be separated from their economic costs. Is it excellent if it costs you several thousand dollars per month? Because that's the real cost: https://www.reddit.com/r/GithubCopilot/comments/1ti3xn0/holy_fuck_how_much_money_was_copilot_losing/
Local models are no match for the ones that are being used for coding, and they still need to be trained. How are you going to keep a local model up to date if you don't have access to a datacenter and billions to burn? It cannot be done locally.
@hopeless and then there's model collapse. This is something you need to be aware if using LLM-based tools: https://www.nature.com/articles/s41586-024-07566-y
Coding models can produce good results because they've been trained on high-quality human produced data, because there were no LLM outputs in public FOSS code. But now that code is being filled with LLM outputs, which will cause the models to inevitably degrade. This is a fact, it's not my opinion, it's how these models work.