oof, classic azure, makes everything so hard
try passing in the service context to the from_defaults() of the sub question engine?
(there was a release 13 hours ago to support passing in the service context here lol)
yeah i implemented it this morning haha
Worked !
But there is another issue now..
I guess it's an indexing error
Because when i index things i have an error because of my Azure Subscription (that i can't get better btw)
However when not doing sub_query there is this warning but i can still query over my docs, which i supposed are incomplete
oof, thats annoying... I can't see the full error, is it a rate limit thing?
I feel we need better rate limiting infrastructure... but in any case, you create an empty index and insert one at a time with a small time.sleep() to slow it down.
This will suuuuck to run, but will avoid the rate limit issue
inserting one node at a time ?
Ok, i managed to find out that i have this indexing error, when it wants to index docs over 10 nodes.
is there a parameter to make it index 10 nodes at a time maybe ?
Yea there's a default batch size for embeddings that generates embeddings for 10 nodes at once I think 🤔
embed_model = OpenAIEmbedding(embed_batch_size=10)
I forget how you are setting up embeddings for azure, but hopefully that works
it was written as 1 at first
but when i put 10 when indexing
InvalidRequestError: Too many inputs. The max number of inputs is 1. We hope to increase the number of inputs per request soon. Please contact us through an Azure support request at: https://go.microsoft.com/fwlink/?linkid=2213926 for further questions.
Lol dang. Guess it has to stay at one 😅
yeaaaah but, it does the job really nicely
Good enough I guess haha 😅
If I find or you too fin a fix on this one let me know ! and thank's again 🙂
just, how would you implement that ?
like a time.sleep() in the indexing phase ?
Yea, but only when that rate limit error comes up. So would need to detect that error and then slow down indexing using time.sleep()
or something like that lol
still get the same error when passing in service_context as a parameter
