DSPy: The framework for programming—not prompting—language models
Python
Fix GEPA docstring format (#8710)
TomuHirata72d2cfc1ecc2
diff --git a/dspy/teleprompt/gepa/gepa.py b/dspy/teleprompt/gepa/gepa.pyindex 97090720e..a4ea0878c 100644--- a/dspy/teleprompt/gepa/gepa.py+++ b/dspy/teleprompt/gepa/gepa.py@@ -197,38 +197,60 @@ class GEPA(Teleprompter):# pareto_frontier is a list of scores, one for each task in the batch.```- Parameters:- - metric: The metric function to use for feedback and evaluation.-- Budget configuration (exactly one of the following must be provided):- - auto: The auto budget to use for the run.- - max_full_evals: The maximum number of full evaluations to perform.- - max_metric_calls: The maximum number of metric calls to perform.-- Reflection based configuration:- - reflection_minibatch_size: The number of examples to use for reflection in a single GEPA step.- - candidate_selection_strategy: The strategy to use for candidate selection. Default is "pareto", which stochastically selects candidates from the Pareto frontier of all validation scores.- - reflection_lm: [Required] The language model to use for reflection. GEPA benefits from a strong reflection model, and you can use `dspy.LM(model='gpt-5', temperature=1.0, max_tokens=32000)` to get a good reflection model.-- Merge-based configuration:- - use_merge: Whether to use merge-based optimization. Default is True.- - max_merge_invocations: The maximum number of merge invocations to perform. Default is 5.-- Evaluation configuration:- - num_threads: The number of threads to use for evaluation with `Evaluate`- - failure_score: The score to assign to failed examples. Default is 0.0.- - perfect_score: The maximum score achievable by the metric. Default is 1.0. Used by GEPA to determine if all examples in a minibatch are perfect.-- Logging configuration:- - log_dir: The directory to save the logs. GEPA saves elaborate logs, along with all the candidate programs, in this directory. Running GEPA with the same `log_dir` will resume the run from the last checkpoint.- - track_stats: Whether to return detailed results and all proposed programs in the `detailed_results` attribute of the optimized program. Default is False.- - use_wandb: Whether to use wandb for logging. Default is False.- - wandb_api_key: The API key to use for wandb. If not provided, wandb will use the API key from the environment variable `WANDB_API_KEY`.- - wandb_init_kwargs: Additional keyword arguments to pass to `wandb.init`.- - track_best_outputs: Whether to track the best outputs on the validation set. track_stats must be True if track_best_outputs is True. `optimized_program.detailed_results.best_outputs_valset` will contain the best outputs for each task in the validation set.-- Reproducibility:- - seed: The random seed to use for reproducibility. Default is 0.+ Args:+ metric: The metric function to use for feedback and evaluation.+ auto: The auto budget to use for the run. Options: "light", "medium", "heavy".+ max_full_evals: The maximum number of full evaluations to perform.+ max_metric_calls: The maximum number of metric calls to perform.+ reflection_minibatch_size: The number of examples to use for reflection in a single GEPA step. Default is 3.+ candidate_selection_strategy: The strategy to use for candidate selection. Default is "pareto",+ which stochastically selects candidates from the Pareto frontier of all validation scores.+ Options: "pareto", "current_best".+ reflection_lm: The language model to use for reflection. Required parameter. GEPA benefits from+ a strong reflection model. Consider using `dspy.LM(model='gpt-5', temperature=1.0, max_tokens=32000)`+ for optimal performance.+ skip_perfect_score: Whether to skip examples with perfect scores during reflection. Default is True.+ add_format_failure_as_feedback: Whether to add format failures as feedback. Default is False.+ use_merge: Whether to use merge-based optimization. Default is True.+ max_merge_invocations: The maximum number of merge invocations to perform. Default is 5.+ num_threads: The number of threads to use for evaluation with `Evaluate`. Optional.+ failure_score: The score to assign to failed examples. Default is 0.0.+ perfect_score: The maximum score achievable by the metric. Default is 1.0. Used by GEPA+ to determine if all examples in a minibatch are perfect.+ log_dir: The directory to save the logs. GEPA saves elaborate logs, along with all candidate+ programs, in this directory. Running GEPA with the same `log_dir` will resume the run+ from the last checkpoint.+ track_stats: Whether to return detailed results and all proposed programs in the `detailed_results`+ attribute of the optimized program. Default is False.+ use_wandb: Whether to use wandb for logging. Default is False.+ wandb_api_key: The API key to use for wandb. If not provided, wandb will use the API key+ from the environment variable `WANDB_API_KEY`.+ wandb_init_kwargs: Additional keyword arguments to pass to `wandb.init`.+ track_best_outputs: Whether to track the best outputs on the validation set. track_stats must+ be True if track_best_outputs is True. The optimized program's `detailed_results.best_outputs_valset`+ will contain the best outputs for each task in the validation set.+ seed: The random seed to use for reproducibility. Default is 0.++ Note:+ Budget Configuration: Exactly one of `auto`, `max_full_evals`, or `max_metric_calls` must be provided.+ The `auto` parameter provides preset configurations: "light" for quick experimentation, "medium" for+ balanced optimization, and "heavy" for thorough optimization.++ Reflection Configuration: The `reflection_lm` parameter is required and should be a strong language model.+ GEPA performs best with models like `dspy.LM(model='gpt-5', temperature=1.0, max_tokens=32000)`.+ The reflection process analyzes failed examples to generate feedback for program improvement.++ Merge Configuration: GEPA can merge successful program variants using `use_merge=True`.+ The `max_merge_invocations` parameter controls how many merge attempts are made during optimization.++ Evaluation Configuration: Use `num_threads` to parallelize evaluation. The `failure_score` and+ `perfect_score` parameters help GEPA understand your metric's range and optimize accordingly.++ Logging Configuration: Set `log_dir` to save detailed logs and enable checkpoint resuming.+ Use `track_stats=True` to access detailed optimization results via the `detailed_results` attribute.+ Enable `use_wandb=True` for experiment tracking and visualization.++ Reproducibility: Set `seed` to ensure consistent results across runs with the same configuration."""def __init__(self,