You should check out LlamaIndex they have pretty good stackable search so you can do something like filter by category or keyword and then rank by similarly. Hybrid search is consistently outperforming keyword or vector only methods. https://gpt-index.readthedocs.io/en/latest/index.html also Anthropics Claude model I find has similar or better performance to GPT4 at lower cost and better speed.
That’s true, they can be a bit abstract which makes them more flexible but less approachable. Cohere also just launched a semantic reranking endpoint which is a bit more clear on how to implement (and quick to layer on to an existing search/filter) https://txt.cohere.com/rerank/ interesting to see LLM providers differentiate on more than just models.
Ah, cool, that's a neat endpoint. Not quite what I needed because my problem is actually translating plain text to a query that works within my taxonomy. Maybe it's something I'll open source ...
You should check out LlamaIndex they have pretty good stackable search so you can do something like filter by category or keyword and then rank by similarly. Hybrid search is consistently outperforming keyword or vector only methods. https://gpt-index.readthedocs.io/en/latest/index.html also Anthropics Claude model I find has similar or better performance to GPT4 at lower cost and better speed.
Will take a look. TBH, I haven't been able to wrap my head around the abstractions on a lot of these projects.
That’s true, they can be a bit abstract which makes them more flexible but less approachable. Cohere also just launched a semantic reranking endpoint which is a bit more clear on how to implement (and quick to layer on to an existing search/filter) https://txt.cohere.com/rerank/ interesting to see LLM providers differentiate on more than just models.
Ah, cool, that's a neat endpoint. Not quite what I needed because my problem is actually translating plain text to a query that works within my taxonomy. Maybe it's something I'll open source ...