3 ms·
What I could see fine-tuning being very useful for is efficiency, either getting GPT-4 Level performance out of a smaller model or pruning GPT-4 for your specif
by tornato7 3y ago
What I could see fine-tuning being very useful for is efficiency, either getting GPT-4 Level performance out of a smaller model or pruning GPT-4 for your specific needs.
After all, if I just want to detect from text what color and brightness the user wants to adjust their lights to, it seems inefficient to use a model that's been trained on all of human knowledge, even if I'm sure it'll work just fine.