Why Am I Getting a 400 Error ('contents.parts must not be empty') with My Fine-Tuned Gemini Model in n8n?

Below is a breakdown of the issues you’re encountering and some pointers for next steps.

---

Error in Fine-Tuned Gemini 1.5 Flash Model

Problem:
The error message indicates that the API call failed with a 400 status, and the specific message:

GenerateContentRequest.system_instruction.parts: contents.parts must not be empty.

What This Means:

• The request sent to Google’s Gemini API for your fine-tuned Gemini 1.5 flash model is missing required content in the parts of the system instruction.
• Essentially, the model is expecting one or more non-empty content parts, but the payload sent from your workflow is empty.

Troubleshooting Suggestions:

1. Check Your Node Configuration:

   • Ensure that you’re passing a non-empty value for the parts that are expected by the fine-tuned model.
   • Verify if your system instruction or prompt is being correctly formatted and attached to the request.

2. Payload Validation:

   • If you’re using expressions or templates in n8n to build the request, confirm that they are returning the expected string value.
   • Test with a simpler known-to-work prompt to isolate whether the issue is with the content generation in the request body.

3. Review Model Tuning Documentation:

   • The documentation on model tuning (e.g., using tunedModels.generateContent) might provide insights into the expected structure of the request payload. It is essential to follow these guidelines to avoid missing required parts.

---

Integrating a Vertex AI Fine-Tuned Gemini 1.5 Pro Model

Integration Possibility:

• n8n supports integration with Google Vertex AI Chat Models via a dedicated node. This node allows you to work with various Google Vertex models, and you can select model names like gemini-1.5-pro-001 (or similar) if your account and project have them enabled.

What to Do Next:

1. Use the Google Vertex Chat Model Node:

   • Check out the Google Vertex Chat Model node documentation for details on the available parameters such as Project ID and Model Name.
   • When setting up the node, make sure you configure the node parameters correctly (such as the project ID, model name, maximum number of tokens, sampling temperature, etc.).

2. Credentials Setup:

   • Ensure that your credentials (using a Google service account) are correctly set up. Refer to the credentials documentation if needed.

3. Testing the Integration:

   • Create a simple workflow that sends a prompt to the Vertex AI Chat Model node. This helps verify that the communication with the Vertex model is correctly established before integrating more complex logic.

---

Summary

• Error Clarification: Your error is due to an empty content part in the system instruction payload for the fine-tuned Gemini 1.5 flash model. You should verify your configuration and ensure that you’re sending a non-empty prompt.
• Vertex AI Integration: Yes, integrating a Vertex AI fine-tuned Gemini 1.5 Pro model is possible within n8n using the appropriate node. Make sure to review the node parameters and credentials setup as described in the documentation.

If you continue to face issues, consider isolating the problematic node by testing with minimal inputs and refer to the respective documentation sections for further configuration details.