Custom Gemini integration

[Anna8888]

How can i integrate a fine tuned gemini model into my appscript project

[Michaelkildery]

Hey

To integrate a fine-tuned Gemini model into your Apps Script project, you'll need to follow these steps:

1. Create a project in the AI Platform:

   • Go to the AI Platform Console and sign in with your Google Cloud Platform (GCP) credentials.
   • Create a new project or select an existing one where you want to deploy your Gemini model.

2. Deploy the fine-tuned Gemini model:

   • Upload your fine-tuned Gemini model files to the AI Platform Model Registry.
   • Create a model deployment using the model files you uploaded.
   • Make sure the model deployment is set to "Online" and has an appropriate endpoint URL.

3. Install the AI Platform Apps Script Library:

   • In your Apps Script project, open the Script Editor.
   • Go to Resources > Libraries > Add Library.
   • Search for and install the "AI Platform Apps Script Library".

4. Access the deployed Gemini model from Apps Script:

   • In your Apps Script code, use the aiplatform.prediction service to make predictions using the deployed Gemini model.
   • Replace <MODEL_NAME> with the name of your deployed model and <ENDPOINT_URL> with the endpoint URL you obtained in step 2.

JavaScript

// Import the AI Platform Apps Script Library
var aiplatform = AppsScript.getAiPlatformService();

// Create a prediction request for the Gemini model
var request = {
  model: {
    name: '<MODEL_NAME>',
    version: 'default'
  },
  inputs: {
    'text': 'This is the text you want to process with the Gemini model'
  }
};

// Send the prediction request and get the response
var response = aiplatform.prediction.predict(request);

// Access the prediction results
var predictions = response.predictions;
console.log(predictions);

This code snippet demonstrates the basic steps of integrating a fine-tuned Gemini model into your Apps Script project. You can further customize the code to fit your specific use case and requirements.

[Anna8888]

Thanks.Any idea why the predict int he following scripts yields the error below?

function askGeminii(inputText) {
  
  const BASE = "https://us-central1-aiplatform.googleapis.com";
  const url = `${BASE}/v1/projects/${PROJECT_ID}/locations/us-central1/endpoints/1645825970269061120:predict`;
  
  const data = {
    "instances": [
      {"content": inputText}
    ]
  };

  const options = {
    method: "post",
    headers: { "Authorization": `Bearer ${ACCESS_TOKEN}`, "Content-Type": "application/json" },
    muteHttpExceptions: true,
    payload: JSON.stringify(data)
  };

  const response = UrlFetchApp.fetch(url, options);
  if (response.getResponseCode() === 200) {
    const jsonResponse = JSON.parse(response.getContentText());
    // Assuming that the model returns an array of predictions
    if (jsonResponse.predictions && jsonResponse.predictions.length > 0) {
      return jsonResponse.predictions[0].generatedText; // Adjust according to the actual key in the response
    }
    return "No valid response from model.";
  }
  return "ERROR: " + response.getContentText(); // Better error handling

}

ERROR: { "error": { "code": 400, "message": "Gemini cannot be accessed through Vertex Predict/RawPredict API. Please follow https://cloud.google.com/vertex-ai/docs/generative-ai/start/quickstarts/quickstart-multimodal for Gemini usage.", "status": "FAILED_PRECONDITION" } }