For other personal thoughts, I think the best thing for a tiny decision model … · @Manjusaka_Lee · TwKit

@Manjusaka_Lee

For other personal thoughts, I think the best thing for a tiny decision model is that we can intergrate it into normal pipleline. CI check/git pre commit check. Something like that. Hers's one case I can share and I'm try to working on this. For now, after vibe coding time, we have a lot of PR waiting for review. 50% of those PR are related to database changes. Some indexes have been added; some queries have been added. The human need to judge something like whether the query hit the index? whether the index adding can block table? Yes, We can 100% to use more powerful LLM model to review those patch. But it's slow and expensive. So if we can use some tiny decision model, we can just use a few of low-cost hardware like NVIDIA T4/L4 to intergrate those with whole org repo. The decision model just need to score about the possibilty the database effective and we can just manually review it when it's necessary. I think this would be awesome(lol Some tiny personal thought(lol

曝光 712 · 评论 2 · 点赞 6 · 书签 2