5 ms·
> that would necessitate constant retraining You wouldn't necessarily need to retrain that frequently. If your model outputs hashes / vectors that can be used
by brainfog 3y ago
> that would necessitate constant retraining
You wouldn't necessarily need to retrain that frequently. If your model outputs hashes / vectors that can be used for searching, you just need to run inference on your new data as it comes in.
- Legend2440 3y agoDefinitely this, embeddings would be the modern approach.
- raylad 3y agoThe "modern" approach sounds like it would be a lot worse than this approach both in terms of input and runtime performance. Trendy ("modern") is not necessarily better.