Skip to content

[BUG] Azure.AI.Projects.OpenAI: uploaded PDF file fails to parse in vector store; MPFD issue? #55350

@trrwilson

Description

@trrwilson

Library name and version

Azure.AI.Projects.OpenAI 1.0.0-beta.4

Describe the bug

There's a combination of .NET-specific exacerbation and service brittleness involved that needs to be root-caused:

Uploading a .pdf file with purpose=assistants succeeds, but then attempting to index it via creating a vector store fails. Fetching the vector store file (/vector_stores/{id}/files) retrieves a last error of "failed to parse."

This doesn't repro against OpenAI's endpoint; it also doesn't repro against Foundry when using a .txt file instead of .pdf. Curl also doesn't appear to reproduce the problem.

  • OpenAI working and Foundry not working implies a service parity issue
  • .NET not working when other languages and curl do implies a .NET-specific issue
  • Previous problems like this have related to multipart/form-data (MPFD) content handling in fields

Expected behavior

If I can upload a file and index it with OpenAI's /files and /vector_stores APIs, I should be able to do it with Foundry's.

Actual behavior

I can't.

Reproduction Steps

See description.

Environment

No response

Metadata

Metadata

Assignees

Labels

AI ProjectsClientThis issue is related to a non-management packageOpenAI

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions