4 results found

1. Feature Request: Add Thinking Budget Support to Firebase AI

   Please add support for a thinking_budget parameter when using Firebase AI (Vertex AI) through Firebase. This would allow developers to control token usage, reasoning depth, and overall latency more precisely. It enables better cost management, improves response time consistency, and ensures a smoother experience in streaming and large-scale applications.

   12 votes

   Vote Vote Vote

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Vote

   We’ll send you updates on this idea

   completed  ·  4 comments  ·  Firebase AI Logic  ·  Delete…  ·  Admin →

2. Grounding

   Enable VertexAI web access for googleSearchRetrieval() and Grounding to work on the FirebaseVertexAI.

   21 votes

   Vote Vote Vote

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Vote

   We’ll send you updates on this idea

   completed  ·  2 comments  ·  Firebase AI Logic  ·  Delete…  ·  Admin →

3. Allow access to Gemini preview models

   Currently the latest stable or other stable Gemini 1.5 models are available only (see https://firebase.google.com/docs/vertex-ai/gemini-models#available-model-names). These lag behind with features such as function calling ability (performance), structured output (according to the feature matrix Gemini 1.5 Flash does not support that but we know it has multiple versions already capable of). Why cannot we access the preview model we'd like?

   4 votes

   Vote Vote Vote

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Vote

   We’ll send you updates on this idea

   completed  ·  3 comments  ·  Firebase AI Logic  ·  Delete…  ·  Admin →

4. Add input and output tokens used and model to the response object.

   One of the very useful things about the Open AI API is that the response includes the amount of input and output tokens used , as well as the model that was hit. I am recording this data per call in my firebase instance for my app along with time of processing to monitor performance. I also add the appropriate cost per model in there and so I dynamically calculate the costs per model call as well. This helps me make business case decisions on how I use the AI. 
   It would be great to have that  available in  the Vertex AI API calls as well. 

   ---

   Is the way to get that now to make a separate call?

   const { totalTokens, totalBillableCharacters } = await model.countTokens([prompt, imagePart]);

   One of the very useful things about the Open AI API is that the response includes the amount of input and output tokens used , as well as the model that was hit. I am recording this data per call in my firebase instance for my app along with time of processing to monitor performance. I also add the appropriate cost per model in there and so I dynamically calculate the costs per model call as well. This helps me make business case decisions on how I use the AI. 
   It would be great to have that  available in  the Vertex AI API… more

   3 votes

   Vote Vote Vote

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Vote

   We’ll send you updates on this idea

   completed  ·  1 comment  ·  Firebase AI Logic  ·  Delete…  ·  Admin →