dspy-finetune-bootstrap

Fine-tune a DSPy program using BootstrapFinetune by providing your teacher DSPy module, training examples, and training configuration. The workflow prepares the teacher, generates traces, fine-tunes a student model, and outputs a finetuned program and saved model path ready for deployment.

Use BootstrapFinetune when you need to reduce inference costs, lower latency, or deploy to resource-constrained environments. It distills a large teacher model into an efficient student model while preserving task performance.

You need a dspy.Module program, a list of dspy.Example training instances, an optional metric callable for evaluation, and train_kwargs dictionary with training parameters. These inputs flow through prepare, trace generation, and fine-tuning phases.

Yes. BootstrapFinetune works with any dspy.Module and list of dspy.Example objects, making it compatible with existing DSPy pipelines and training data without requiring format conversion.

BootstrapFinetune outputs a finetuned dspy.Module ready for inference and a model_path string pointing to saved weights that you can load and deploy in production environments.

Yes. By distilling a large teacher model into fine-tuned weights, BootstrapFinetune eliminates reliance on expensive API-only inference and enables cost-effective self-hosted deployment.