Gemini Batch API not accepting files

Google Cloud Build with AI AI APIs
1

When I create a batch and I just send text I have zero issues, but once I send pics it doesn’t work. Here is the JSONL

I upload them using the file API. I’ve spent a ton of time researching this and I can’t figure out why I’m getting the error. I’ve tried with 2.5 flash and pro and I get the same error.

Found an error in result for key ‘request-0’: Request contains an invalid argument.

{“key”: “request-0”, “request”: {“contents”: [{“role”: “user”, “parts”: [{“text”: “prompt is here”]}"}, {“file_data”: {“file_uri”: “files/9yex9nx60jy3”, “mime_type”: “image/jpeg”}}]}], “generation_config”: {“response_mime_type”: “application/json”}}}

2

Hello @Cute_Panda :panda:,

Your question is interesting, so I’ve checked the documentation out, and it seems that the Input Data Requirements for file list indicates that you need to use a GCS path.

If you plan to use the Google Cloud console to get batch inferences, paste your file list directly into the Google Cloud console. Otherwise, save the list in a Cloud Storage bucket.

You can also check how to Get Batch Inferences from an image classification model.