Find answers from the community

Updated 2 years ago

I am also experimenting with whatsapp

I am also experimenting with whatsapp embeddings...is there a recommended better approach than GPTSimpleVectorIndex for that data structure? I tried GPTTreeIndex but the instance just hangs...not sure if I was barking up the wrong tree with my approach

9 comments

jjerryjliu0

re: embedding support, you should probably stick with a vector index for now 🙂 i would check out the composability feature if you want to design graphs over your data to more precisely feed in context during query-time

ssmokeoX

thanks boss-man! Yeah the whatsapp queries were odd, it couldn't identify one of the two people in the conversation or any of the content really. I will dive deeper into composability to get these fine tuned! :goodjob:

jjerryjliu0

ofc, np. @smokeoX when you say whatsapp embeddings do you mean using openai embeddings on whatsapp content?

ssmokeoX

using the whatsapp loader on llamahub, which is loading a chat history as a .txt

ssmokeoX

and using GPTSimpleVectorIndex on this

Plain Text

WhatsappChatLoader = download_loader("WhatsappChatLoader")
path = "_chat.txt"
loader = WhatsappChatLoader(path=path)
documents = loader.load_data()

ssmokeoX

and curious if you've encountered the script hanging when using the other index types @jerryjliu0?

jjerryjliu0

Ooh I see.

jjerryjliu0

the script might be hanging due to index construction being super slow. we're hoping to speed some of this up soon by adding async to index construction

jjerryjliu0

to start with i'd start with the vector indices though

Add a reply