this video is very helpful, thank you. I was upset because of the vector dbs. I will check if I will be able to use the cloud version and buy the paid plan if I am successful
OpenAi embeddings only output Cosine. Milvus has GPU support but only for L2 and IP (not for Cosine) from what I understand? Is there another provider that can output L2 or IP other than openAI? Or does that not make sense to use a GPU for text, is that more intended for images and audio?
@@MilvusVectorDatabase yes but with cosine that would be without GPU support right? As GPU support is only for L2 and IP (not for Cosine)? If that is correct I would need another embeddings system but all I find for text are cosine.
if you want true separation you'll probably need logical partitions, for functional separation i would suggest just adding the username as a metadata field and filtering on it
@@yujiantang Yeah i made something more crazy that i edited langchain and added my own userid field then released my own version😂, i tried adding metadata field but i couldn't do it manually so i did some wierd stuff till it worked.
this video is very helpful, thank you. I was upset because of the vector dbs. I will check if I will be able to use the cloud version and buy the paid plan if I am successful
glad you found it helpful!
Thanks for the great information! What does you mean by “to manage utility”?
OpenAi embeddings only output Cosine. Milvus has GPU support but only for L2 and IP (not for Cosine) from what I understand? Is there another provider that can output L2 or IP other than openAI? Or does that not make sense to use a GPU for text, is that more intended for images and audio?
Milvus 2.3 has cosine similarity metrics
This is a good read zilliz.com/learn/choosing-right-vector-index-for-your-project
@@MilvusVectorDatabase yes but with cosine that would be without GPU support right? As GPU support is only for L2 and IP (not for Cosine)? If that is correct I would need another embeddings system but all I find for text are cosine.
How would you replace ConversationChain with LCEL ? Seems impossible
im using it and its amazing , but im having trouble seperating the memory for each user , any ideas ?
if you want true separation you'll probably need logical partitions, for functional separation i would suggest just adding the username as a metadata field and filtering on it
@@yujiantang Yeah i made something more crazy that i edited langchain and added my own userid field then released my own version😂, i tried adding metadata field but i couldn't do it manually so i did some wierd stuff till it worked.
@@Cuzinz oh but you got it working? i'm curious to see what you did if you wanna share :)
It would be nice if the link to the github repo with the code was provided so I could work through it while watching.