Repo or source
https://github.com/stanfordnlp/dspy
What to explain
The first optimizer: BootstrapFewShot.compile (dspy/teleprompt/bootstrap.py) makes a reset_copy student and a deepcopy teacher (optionally seeded with LabeledFewShot demos), runs the teacher on the trainset while dspy.settings.trace records every predictor's (inputs, outputs), keeps only runs the metric accepts, and installs the surviving per-predictor steps as demos on the matching student predictors via named_predictors. Cover BootstrapFewShotWithRandomSearch (random_search.py) as the wrapper that tries many bootstrapped candidates and keeps the best-scoring one. The idea to land: the program teaches itself few-shot examples from its own successful executions.
This book is part of the series "DSPY" (book 6 of 8) — pass --series "DSPY" --series-order 6 to video.mjs.
Cover title
Bootstrap
Subtitle (optional)
Teacher, Student, and the Trace
Cover animal (optional)
kangaroo
Accent color (optional hex)
#f97316
Model
(claude-fable-5)
Repo or source
https://github.com/stanfordnlp/dspy
What to explain
The first optimizer: BootstrapFewShot.compile (dspy/teleprompt/bootstrap.py) makes a reset_copy student and a deepcopy teacher (optionally seeded with LabeledFewShot demos), runs the teacher on the trainset while dspy.settings.trace records every predictor's (inputs, outputs), keeps only runs the metric accepts, and installs the surviving per-predictor steps as demos on the matching student predictors via named_predictors. Cover BootstrapFewShotWithRandomSearch (random_search.py) as the wrapper that tries many bootstrapped candidates and keeps the best-scoring one. The idea to land: the program teaches itself few-shot examples from its own successful executions.
This book is part of the series "DSPY" (book 6 of 8) — pass --series "DSPY" --series-order 6 to video.mjs.
Cover title
Bootstrap
Subtitle (optional)
Teacher, Student, and the Trace
Cover animal (optional)
kangaroo
Accent color (optional hex)
#f97316
Model
(claude-fable-5)