Get hands-on experience with 20+ free Google Cloud products and $300 in free credit for new customers.

Hyperparameters for GenAI

Hello! I hope someone can shed some light on an issue i'm having.

I am having an issue with consistency when switching from Vertex API to GenAI API.

Having identical model and hyperparamters, for the same prompt generated via:
Vertex API: I get consistent results all the time.
In GenAI API: Wildly varying results.

Extracted relevant code:

For VertexAI I have:

vertexAi = new VertexAI({
project: this.PROJECT,
location: "europe-west1",
});
 
const model = this.vertexAi.getGenerativeModel({
model: "gemini-2.0-flash-001",
systemInstruction: systemInstruction,
generationConfig: {
    temperature: 0.0,
    topP: 0.1,
    topK: 10,
    seed: 1337,
    responseMimeType: "application/json",
    responseSchema: schema,
},
});
 
aiResponse = await model.generateContent({
    contents: [
        {role: "user", parts: [{text: JSON.stringify(lastUserMessage)}]},
    ],
});

____

For GenAI API I have:

genAI = new GoogleGenerativeAI(config.apiKey);
 
const model = this.genAI.getGenerativeModel({
    model: "gemini-2.0-flash-001",
    systemInstruction: systemInstruction
});
 
const aiResponse = await model.generateContent({
contents: [
{role: "user", parts: [{text: JSON.stringify(lastUserMessage)}]},
],
generationConfig: {
    temperature: 0.0,
    topP: 0.1,
    topK: 10,
    seed: 1337,
    responseMimeType: "application/json",
    responseSchema: schema,
},
});
 
Basically just the generation config moved as per the documentation.

Anyone have any insights? I'm pretty sure temperature etc is not being taken into account in my GenAI version.
 
0 0 64
0 REPLIES 0