Picking the right 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴 𝗺𝗼𝗱𝗲𝗹 is crucial.

Can't go to warp without a good vectorization engine!

Qdrant supports dense, sparse, and 𝗺𝘂𝗹𝘁𝗶𝘃𝗲𝗰𝘁𝗼𝗿 embeddings. However, the vectors' performance largely depends on the embedding model and its ability to capture meaning.

What's the best model for you?

1️⃣ Look at benchmarks like 𝗠𝗧𝗘𝗕 or 𝗕𝗘𝗜𝗥. But remember - the top model might not perform the same in your setup. You need to experiment!

2️⃣ Ask the community & experts. On 𝗤𝗱𝗿𝗮𝗻𝘁'𝘀 𝗗𝗶𝘀𝗰𝗼𝗿𝗱, I am seeing 𝗻𝗼𝗺𝗶𝗰-𝗲𝗺𝗯𝗲𝗱-𝘁𝗲𝘅𝘁 perform better than OpenAI’s 𝘁𝗲𝘅𝘁-𝘀𝗺𝗮𝗹𝗹-𝟯 for text retrieval. Try to learn from the experiences of others.

💬 What’s your go-to embedding model? Are you using 𝘦5, 𝘯𝘰𝘮𝘪𝘤, 𝘖𝘱𝘦𝘯𝘈𝘐’𝘴 𝘦𝘮𝘣𝘦𝘥𝘥𝘪𝘯𝘨𝘴, 𝘰𝘳 𝘴𝘰𝘮𝘦𝘵𝘩𝘪𝘯𝘨 𝘦𝘭𝘴𝘦? Let me know in the comments!

P.S. 𝗖𝗼𝗵𝗲𝗿𝗲 𝗘𝗺𝗯𝗲𝗱 𝟯 just dropped—a multimodal model that handles complex docs and catalogs better than CLIP. Worth a try?