something went wrong
johndpope opened this issue · comments
Happening the same to me with certain PDFs files. Actually, I did a test and divided that big PDF into one-page size PDFs and I tried with each of them and I didn't have any issues. Unfortunately, the description of the Pinecone issue is very poor.
I think there's no attribute message e.message - the error is "e" object.
Thanks @Klaudioz @johndpope do you know a way to reproduce this consistently? Happy to handle this case to provide a better error feedback.
Please send me a PM on Twitter (https://twitter.com/klaudioz) to share the file with you.
This error is usually because the the client is not able to find the index created under the account, can you double check the followings:
- index name is
book-gpt
- index region is in
us-west1-gcp
(I will make this configurable in the future)
thanks for troubleshooting - after updating env to us-east - I finally get upload success callback.
This popped up on my radar - maybe useful
https://github.com/johndpope/PatrikZeros-ChatGPT-API-UI
us-west1-gcp
thanks for troubleshooting - after updating env to us-east - I finally get upload success callback. This popped up on my radar - maybe useful https://github.com/johndpope/PatrikZeros-ChatGPT-API-UI
Is a paid plan required to change the environment? I can't figure out how to do this from the web UI
Nevermind, I see this was hard-coded in. Finally got it working on my end!
To anybody seeing this, just make sure you change the environment to us-east1-gcp
or whatever your specific region is in both pages/api/ingest.ts
AND pages/api/chat.ts
Great work @fraserxu and thanks for your help troubleshooting as well @johndpope!
@Klaudioz I tried your provided pdf and can reproduce the issue. I was able to get better error response after upgraded the pinecone
npm package thanks to @rschwabco 's recent work on better error handling.
[Error: PineconeClient: Error calling upsert: Error: Your request is larger than the maximum supported size - 2MB. Please try to reduce your batch size.]
A short term fix is maybe try to cut the pdf manually by yourself. I'll be looking at if I can do this with code or maybe adjust batch size or something in the next few days.
Thanks for your answer. That limitation sounds weird to me. I've uploaded bigger files in the past without any issues.
I'd really like to know where that limitation is mentioned in Pinecone's documentation.
I'm going to update the Pinecone package right now!!
@johndpope @Klaudioz while we waiting for the issue being clarified by the upstream, I've implemented a simple chunk function which would write the doc in to pinecone with small chunks.
I tested with the PDF @Klaudioz provided and can query fine. Please let me know if it works for you, cheers.