supabase-community / chatgpt-your-files

Production-ready MVP for securely chatting with your documents using pgvector
https://youtu.be/ibzlEQmgPPY
325 stars 113 forks source link

Partial NULL embeddings #35

Open Nishanth-Gobi opened 3 months ago

Nishanth-Gobi commented 3 months ago

Bug report

Describe the bug

On file uploads, the embedding function fails to create vectors for some of the chunks with the below message,

InvalidWorkerCreation: worker did not respond in time
    at async UserWorker.create (ext:sb_user_workers/user_workers.js:155:15)
    at async Object.handler (file:///home/deno/main/index.ts:153:22)
    at async respond (ext:sb_core_main_js/js/http.js:162:14) {
  name: "InvalidWorkerCreation"
}

I am using the new Supabase.ai.Session() API and am on Edge v1.53.3, but it still happens sometimes at random.

To Reproduce

Steps to reproduce the behavior, please provide code snippets or a repository:

  1. Run main locally
  2. Upload a couple of markdown files one after the other

Expected behavior

All chunks to have their corresponding vectors.

Screenshots

image

System information

Additional context

@gregnr this seems to be different from what you mentioned could be the reason for partial NULLs in #30

gregnr commented 3 months ago

Very strange @Nishanth-Gobi, thanks for reporting. You're the first I've seen that has experienced this with the new Supabase.ai API. Would you mind submitting an issue on the edge-runtime repo with this error (and link back to this issue)?

landmann commented 6 days ago

Ahh ok I found the issue @Nishanth-Gobi - you have a bug in your functions/embed/index.ts file. Mine was that I didn't generate the typescript type into _lib/database.ts