If you’re on older Gemini models, it’s time to start planning your move to the latest models for significant performance improvements and new capabilities.
For a smooth migration, here are a few things to keep in mind:
Before you begin: Be aware of model retirement dates, check for location availability, and understand pricing differences. See links below.
Upgrade your code: Consider switching to the Gen AI SDK, as new capabilities are only available there.
Evaluate performance: Repeat your evaluations and perform offline testing to ensure the new model’s output quality is at least as good as the old one. The Vertex AI GenAI evaluation service has been improved to make model migrations much easier.
Load test: If your application has specific throughput requirements, conduct load testing to confirm the new model version can handle the demand.
Tune as needed: If you see a performance drop, iteratively engineer your prompts or fine-tune the model to match the old model’s performance.
Deploy to production: Once the new model meets or exceeds performance, you’re ready for a production rollout.
For more details, check out the docs:
✦ Migration guide: https://cloud.google.com/vertex-ai/generative-ai/docs/migrate-to-v2
✦ Model lifecycle: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions