Flip insights into product excellence
Each consumer interplay is an opportunity to enhance your product and the mannequin’s means to ship higher responses. You possibly can export your logs as particular datasets (in CSV or JSONL format) for testing and offline analysis. By figuring out examples in your logs the place high quality and efficiency dipped (or excelled), you’ll be able to construct a dependable and reproducible baseline of anticipated outcomes.
You should use these datasets for immediate refinement, efficiency monitoring and extra. For instance, you should utilize the Gemini Batch API to run batch evaluations utilizing datasets constructed up over time, see the Datasets Cookbook for an instance. This lets you check the modifications to your Gemini mannequin choice or utility logic earlier than you deploy them to customers.
You even have the choice to share particular datasets with Google to offer suggestions on end-to-end mannequin habits on your particular use case. Shared datasets can be used to enhance and develop Google services and products, together with bettering and coaching our fashions.
