from Hacker News

Ask HN: Best tools to create LLM prodcuts?

by funerr on 5/26/23, 8:11 AM with 2 comments

I'd like to maximize DX. I'm looking for reasonable pricing for embedding databases, fine-tuning models and chatbot interfaces. What's the best game in town for those?

  • by tikkun on 5/26/23, 1:36 PM

    1. Embedding databases: pgvector, storing embeddings in plain json (if you don't have too many), chroma, weaviate, pinecone

    2. Fine-tuning: The better a model is, the more it's worth fine tuning it. GPT-4 is the best model by far. To fine tune it you need to be an OpenAI foundry customer, 6 figures plus spend. Otherwise, any GPU compute provider can allow you to fine tune an open model

    3. Interfaces: Not quite sure what you mean here, but Vercel has some templates/open source examples

  • by kacperlukawski on 5/26/23, 1:45 PM

    Definitely Qdrant is a great option. It's an open source vector DB, so you can experiment locally without any cost, but if you prefer managed solution Qdrant Cloud is an option: https://cloud.qdrant.io/