4 ms·
I'd rather stick with InstructEmbedding instead of pandering to flavour of the month LLM. That way I keep my key components insulated from drastic changes
by villgax 3y ago
I'd rather stick with InstructEmbedding instead of pandering to flavour of the month LLM. That way I keep my key components insulated from drastic changes
- villgax 3y agoLLM inputs are the worst candidates for caching. The only place where caching might make sense if you have a public facing service & have coupled it with a vector cache instead of a typical word for word caching of the prompt
- eigenvalue 3y agoIt’s useful if you might submit the same document or edited versions of the same document with a lot of overlap.