dspy

Define a metric function and training examples, then compile your module with an optimizer like BootstrapFewShot or MIPRO. The optimizer generates few-shot demonstrations or searches for better instructions, returning an improved module without manual prompt editing.

Predict makes a direct LM call for simple tasks, while ChainOfThought generates reasoning steps before the answer. ChainOfThought is roughly twice as slow but significantly more accurate on math and logical reasoning tasks.

DSPy offers automatic data-driven prompt optimization, typed signatures, and higher modularity, while LangChain provides manual chains with many existing tool integrations. Choose DSPy when you have training data and need systematic improvement across different LMs.

Yes, DSPy supports local models through dspy.OllamaLocal by pointing to a local base URL such as http://localhost:11434. It also supports OpenAI, Anthropic Claude, and configuring multiple models for different pipeline stages.

BootstrapFewShot works with 10-50 examples, MIPRO benefits from 50-200 examples plus a separate validation set, and BootstrapFinetune recommends 100 or more examples. Insufficient data is a common cause of poor optimization results.

Avoid DSPy for quick one-off prototypes where manual prompting suffices, or when you lack any training data for optimization. It also has a medium-to-high learning curve compared to simple prompt templates.