Ground Model Responses Error

[tlloyds]

I've been playing recently with very high level grounding of model responses in a datastore by creating a search app in Agent Builder and creating a datastore there that points to my data in a bucket. I started with .txt files and it seemed to work fine where I could just go to model garden, go into gemini 1.5, then under advanced point to the datastore. I'd ask questions about the data and get good answers. Tonight I added a bucket of .pdf files and did the same thing, but every prompt results in the following error:

Failed to submit prompt

Error message: "Request contains an invalid argument."

Status: 400 Error code: 3

Request ID: 1810478650213582325

I've tried every variation of prompt imaginable, and I tried pointing to the datastore that worked the other night for me and now but now can't get a response. The search itself works fine on the data if I just go into the search app and query either datastore, but now I can't get the generic GUI gemini interface to return a result. Is there anything I can try or do by way of next step? It seems odd that it isn't working tonight on the older datastore. The only real configuration here is pointing to the datastore in the advanced options. Thanks in advance for any help!

[ruthseki]

Hi @tlloyds,

Welcome to Google Cloud Community!

For error code 400 with invalid argument status,  here are some workarounds that you may try:

• Check the API reference for request format, examples, and supported versions. Using features from a newer API version with an older endpoint can cause errors.
• Look for any error messages via Logs Explorer . These might provide clues as to why it isn't working.
• You may also try to restart the Gemini model and any related services in the Agent Builder and see if it works. Please try using a different supported browser such as Google Chrome, Safari, Firefox or Edge to isolate any browser-related issues. 

If you’re still encountering the issue, contact Google Cloud Support for assistance.  Provide the complete error message, including the request ID, for the best assistance.

I hope the above information is helpful.