dspy

Automate prompt optimization by defining declarative LM pipelines and using teleprompters with data-driven metrics to generate and evaluate instructions. You provide a labeled trainset and an evaluation metric to improve model outputs.

Declarative programming for language models defines tasks and workflows without imperative prompts, enabling reuse and safety. It shifts focus from manual prompt engineering to modular reasoning definitions optimized automatically by teleprompters.

Optimize few-shot demonstrations for RAG workflows using optimizers like BootstrapFewShot or MIPRO. These teleprompters generate and evaluate prompts against a defined metric using a labeled trainset to improve multi-stage QA outputs.

Yes, automated prompt optimization supports multiple LM providers like OpenAI and Claude. You can optimize multi-stage QA and RAG pipelines across different providers using data-driven metrics and teleprompters.

Yes, you need a labeled trainset and a defined evaluation metric to run automated prompt optimization. A validation set is optional but recommended to support optimizers like BootstrapFewShot, MIPRO, and KNNFewShot.

Teleprompter-driven optimization requires structured labeled data and a defined evaluation metric to function. Without a clear metric or sufficient trainset data, optimizers like BootstrapFewShot or COPRO cannot effectively evaluate and improve prompt quality.