DSPy: The framework for programming—not prompting—language models
Python
Change the output interface of evaluate (#8003)
TomuHiratad52f7a5f80fc
diff --git a/docs/docs/tutorials/agents/index.ipynb b/docs/docs/tutorials/agents/index.ipynbindex 493ce4a05..d2ce3c71f 100644--- a/docs/docs/tutorials/agents/index.ipynb+++ b/docs/docs/tutorials/agents/index.ipynb@@ -500,23 +500,20 @@" metric=top5_recall,\n"," num_threads=16,\n"," display_progress=True,\n",- " # To record the outputs and detailed scores to MLflow\n",- " return_all_scores=True,\n",- " return_outputs=True,\n"," )\n","\n"," # Evaluate the program as usual\n",- " aggregated_score, outputs, all_scores = evaluate(cot)\n",+ " result = evaluate(cot)\n","\n"," # Log the aggregated score\n",- " mlflow.log_metric(\"top5_recall\", aggregated_score)\n",+ " mlflow.log_metric(\"top5_recall\", result.score)\n"," # Log the detailed evaluation results as a table\n"," mlflow.log_table(\n"," {\n"," \"Claim\": [example.claim for example in eval_set],\n"," \"Expected Titles\": [example.titles for example in eval_set],\n",- " \"Predicted Titles\": outputs,\n",- " \"Top 5 Recall\": all_scores,\n",+ " \"Predicted Titles\": [output[1] for output in result.results],\n",+ " \"Top 5 Recall\": [output[2] for output in result.results],\n"," },\n"," artifact_file=\"eval_results.json\",\n"," )\n",diff --git a/docs/docs/tutorials/classification_finetuning/index.ipynb b/docs/docs/tutorials/classification_finetuning/index.ipynbindex b318e4725..80f949a43 100644--- a/docs/docs/tutorials/classification_finetuning/index.ipynb+++ b/docs/docs/tutorials/classification_finetuning/index.ipynb@@ -568,23 +568,20 @@" metric=extraction_correctness_metric,\n"," num_threads=16,\n"," display_progress=True,\n",- " # To record the outputs and detailed scores to MLflow\n",- " return_all_scores=True,\n",- " return_outputs=True,\n"," )\n","\n"," # Evaluate the program as usual\n",- " aggregated_score, outputs, all_scores = evaluate_correctness(people_extractor)\n",+ " result = evaluate_correctness(people_extractor)\n","\n"," # Log the aggregated score\n",- " mlflow.log_metric(\"exact_match\", aggregated_score)\n",+ " mlflow.log_metric(\"exact_match\", result.score)\n"," # Log the detailed evaluation results as a table\n"," mlflow.log_table(\n"," {\n"," \"Text\": [example.text for example in devset],\n"," \"Expected\": [example.example_label for example in devset],\n",- " \"Predicted\": outputs,\n",- " \"Exact match\": all_scores,\n",+ " \"Predicted\": [output[1] for output in result.results],\n",+ " \"Exact match\": [output[2] for output in result.results],\n"," },\n"," artifact_file=\"eval_results.json\",\n"," )\n",diff --git a/docs/docs/tutorials/entity_extraction/index.ipynb b/docs/docs/tutorials/entity_extraction/index.ipynbindex d67bfc193..f948e9ef8 100644--- a/docs/docs/tutorials/entity_extraction/index.ipynb+++ b/docs/docs/tutorials/entity_extraction/index.ipynb@@ -514,23 +514,20 @@" metric=extraction_correctness_metric,\n"," num_threads=24,\n"," display_progress=True,\n",- " # To record the outputs and detailed scores to MLflow\n",- " return_all_scores=True,\n",- " return_outputs=True,\n"," )\n","\n"," # Evaluate the program as usual\n",- " aggregated_score, outputs, all_scores = evaluate_correctness(people_extractor)\n",+ " result = evaluate_correctness(people_extractor)\n","\n"," # Log the aggregated score\n",- " mlflow.log_metric(\"exact_match\", aggregated_score)\n",+ " mlflow.log_metric(\"exact_match\", result.score)\n"," # Log the detailed evaluation results as a table\n"," mlflow.log_table(\n"," {\n"," \"Tokens\": [example.tokens for example in test_set],\n"," \"Expected\": [example.expected_extracted_people for example in test_set],\n",- " \"Predicted\": outputs,\n",- " \"Exact match\": all_scores,\n",+ " \"Predicted\": [output[1] for output in result.results],\n",+ " \"Exact match\": [output[2] for output in result.results],\n"," },\n"," artifact_file=\"eval_results.json\",\n"," )\n",diff --git a/docs/docs/tutorials/math/index.ipynb b/docs/docs/tutorials/math/index.ipynbindex a0f70065c..10de528de 100644--- a/docs/docs/tutorials/math/index.ipynb+++ b/docs/docs/tutorials/math/index.ipynb@@ -369,21 +369,21 @@"\n","# Start an MLflow Run to record the evaluation\n","with mlflow.start_run(run_name=\"math_evaluation\"):\n",- " kwargs = dict(num_threads=THREADS, display_progress=True, return_all_scores=True, return_outputs=True)\n",+ " kwargs = dict(num_threads=THREADS, display_progress=True)\n"," evaluate = dspy.Evaluate(devset=dataset.dev, metric=dataset.metric, **kwargs)\n","\n"," # Evaluate the program as usual\n",- " aggregated_score, outputs, all_scores = evaluate(module)\n",+ " result = evaluate(module)\n","\n"," # Log the aggregated score\n",- " mlflow.log_metric(\"correctness\", aggregated_score)\n",+ " mlflow.log_metric(\"correctness\", result.score)\n"," # Log the detailed evaluation results as a table\n"," mlflow.log_table(\n"," {\n"," \"Question\": [example.question for example in dataset.dev],\n"," \"Gold Answer\": [example.answer for example in dataset.dev],\n",- " \"Predicted Answer\": outputs,\n",- " \"Correctness\": all_scores,\n",+ " \"Predicted Answer\": [output[1] for output in result.results],\n",+ " \"Correctness\": [output[2] for output in result.results],\n"," },\n"," artifact_file=\"eval_results.json\",\n"," )\n",diff --git a/docs/docs/tutorials/multihop_search/index.ipynb b/docs/docs/tutorials/multihop_search/index.ipynbindex b769b02e7..00ed91e82 100644--- a/docs/docs/tutorials/multihop_search/index.ipynb+++ b/docs/docs/tutorials/multihop_search/index.ipynb@@ -534,23 +534,20 @@" metric=top5_recall,\n"," num_threads=16,\n"," display_progress=True,\n",- " # To record the outputs and detailed scores to MLflow\n",- " return_all_scores=True,\n",- " return_outputs=True,\n"," )\n","\n"," # Evaluate the program as usual\n",- " aggregated_score, outputs, all_scores = evaluate(Hop())\n",+ " result = evaluate(Hop())\n","\n"," # Log the aggregated score\n",- " mlflow.log_metric(\"top5_recall\", aggregated_score)\n",+ " mlflow.log_metric(\"top5_recall\", result.score)\n"," # Log the detailed evaluation results as a table\n"," mlflow.log_table(\n"," {\n"," \"Claim\": [example.claim for example in eval_set],\n"," \"Expected Titles\": [example.titles for example in eval_set],\n",- " \"Predicted Titles\": outputs,\n",- " \"Top 5 Recall\": all_scores,\n",+ " \"Predicted Titles\": [output[1] for output in result.results],\n",+ " \"Top 5 Recall\": [output[2] for output in result.results],\n"," },\n"," artifact_file=\"eval_results.json\",\n"," )\n",diff --git a/docs/docs/tutorials/rag/index.ipynb b/docs/docs/tutorials/rag/index.ipynbindex b3d29e83e..0f4a241a1 100644--- a/docs/docs/tutorials/rag/index.ipynb+++ b/docs/docs/tutorials/rag/index.ipynb@@ -731,24 +731,21 @@" metric=metric,\n"," num_threads=24,\n"," display_progress=True,\n",- " # To record the outputs and detailed scores to MLflow\n",- " return_all_scores=True,\n",- " return_outputs=True,\n"," )\n","\n"," # Evaluate the program as usual\n",- " aggregated_score, outputs, all_scores = evaluate(cot)\n",+ " result = evaluate(cot)\n","\n","\n"," # Log the aggregated score\n",- " mlflow.log_metric(\"semantic_f1_score\", aggregated_score)\n",+ " mlflow.log_metric(\"semantic_f1_score\", result.score)\n"," # Log the detailed evaluation results as a table\n"," mlflow.log_table(\n"," {\n"," \"Question\": [example.question for example in eval_set],\n"," \"Gold Response\": [example.response for example in eval_set],\n",- " \"Predicted Response\": outputs,\n",- " \"Semantic F1 Score\": all_scores,\n",+ " \"Predicted Response\": [output[1] for output in result.results],\n",+ " \"Semantic F1 Score\": [output[2] for output in result.results],\n"," },\n"," artifact_file=\"eval_results.json\",\n"," )\n",@@ -1471,6 +1468,11 @@"\n","The first step is to look at your system outputs, which will allow you to identify the sources of lower performance if any. While doing all of this, make sure you continue to refine your metric, e.g. by optimizing against your judgments, and to collect more (or more realistic) data, e.g. from related domains or from putting a demo of your system in front of users."]+ },+ {+ "cell_type": "markdown",+ "metadata": {},+ "source": []}],"metadata": {diff --git a/dspy/evaluate/evaluate.py b/dspy/evaluate/evaluate.pyindex ea86c0c04..ee05ff7df 100644--- a/dspy/evaluate/evaluate.py+++ b/dspy/evaluate/evaluate.py@@ -9,6 +9,7 @@ if TYPE_CHECKING:import tqdmimport dspy+from dspy.primitives.prediction import Predictionfrom dspy.utils.callback import with_callbacksfrom dspy.utils.parallelizer import ParallelExecutor@@ -42,6 +43,21 @@ except ImportError:logger = logging.getLogger(__name__)+class EvaluationResult(Prediction):+ """+ A class that represents the result of an evaluation.+ It is a subclass of `dspy.Prediction` that contains the following fields++ - score: An float value (e.g., 67.30) representing the overall performance+ - results: a list of (example, prediction, score) tuples for each example in devset+ """+ def __init__(self, score: float, results: list[Tuple["dspy.Example", "dspy.Example", Any]]):+ super().__init__(score=score, results=results)++ def __repr__(self):+ return f"EvaluationResult(score={self.score}, results=<list of {len(self.results)} results>)"++class Evaluate:"""DSPy Evaluate class.@@ -58,8 +74,6 @@ class Evaluate:display_progress: bool = False,display_table: Union[bool, int] = False,max_errors: Optional[int] = None,- return_all_scores: bool = False,- return_outputs: bool = False,provide_traceback: Optional[bool] = None,failure_score: float = 0.0,**kwargs,@@ -74,8 +88,6 @@ class Evaluate:If a number is passed, the evaluation results will be truncated to that number before displayed.max_errors (Optional[int]): The maximum number of errors to allow beforestopping evaluation. If ``None``, inherits from ``dspy.settings.max_errors``.- return_all_scores (bool): Whether to return scores for every data record in `devset`.- return_outputs (bool): Whether to return the dspy program's outputs for every data in `devset`.provide_traceback (Optional[bool]): Whether to provide traceback information during evaluation.failure_score (float): The default score to use if evaluation fails due to an exception."""@@ -85,8 +97,6 @@ class Evaluate:self.display_progress = display_progressself.display_table = display_tableself.max_errors = max_errors- self.return_all_scores = return_all_scores- self.return_outputs = return_outputsself.provide_traceback = provide_tracebackself.failure_score = failure_score@@ -99,10 +109,8 @@ class Evaluate:num_threads: Optional[int] = None,display_progress: Optional[bool] = None,display_table: Optional[Union[bool, int]] = None,- return_all_scores: Optional[bool] = None,- return_outputs: Optional[bool] = None,callback_metadata: Optional[dict[str, Any]] = None,- ):+ ) -> EvaluationResult:"""Args:program (dspy.Module): The DSPy program to evaluate.@@ -114,36 +122,20 @@ class Evaluate:`self.display_progress`.display_table (Union[bool, int]): Whether to display the evaluation results in a table. if not provided, use`self.display_table`. If a number is passed, the evaluation results will be truncated to that number before displayed.- return_all_scores (bool): Whether to return scores for every data record in `devset`. if not provided,- use `self.return_all_scores`.- return_outputs (bool): Whether to return the dspy program's outputs for every data in `devset`. if not- provided, use `self.return_outputs`.callback_metadata (dict): Metadata to be used for evaluate callback handlers.Returns:- The evaluation results are returned in different formats based on the flags:-- - Base return: A float percentage score (e.g., 67.30) representing overall performance-- - With `return_all_scores=True`:- Returns (overall_score, individual_scores) where individual_scores is a list of- float scores for each example in devset-- - With `return_outputs=True`:- Returns (overall_score, result_triples) where result_triples is a list of- (example, prediction, score) tuples for each example in devset-- - With both flags=True:- Returns (overall_score, result_triples, individual_scores)-+ The evaluation results are returned as a dspy.EvaluationResult object containing the following attributes:++ - score: A float percentage score (e.g., 67.30) representing overall performance++ - results: a list of (example, prediction, score) tuples for each example in devset"""metric = metric if metric is not None else self.metricdevset = devset if devset is not None else self.devsetnum_threads = num_threads if num_threads is not None else self.num_threadsdisplay_progress = display_progress if display_progress is not None else self.display_progressdisplay_table = display_table if display_table is not None else self.display_table- return_all_scores = return_all_scores if return_all_scores is not None else self.return_all_scores- return_outputs = return_outputs if return_outputs is not None else self.return_outputsif callback_metadata:logger.debug(f"Evaluate is called with callback metadata: {callback_metadata}")@@ -194,14 +186,11 @@ class Evaluate:else:logger.warning("Skipping table display since `pandas` is not installed.")- if return_all_scores and return_outputs:- return round(100 * ncorrect / ntotal, 2), results, [score for *_, score in results]- if return_all_scores:- return round(100 * ncorrect / ntotal, 2), [score for *_, score in results]- if return_outputs:- return round(100 * ncorrect / ntotal, 2), results+ return EvaluationResult(+ score=round(100 * ncorrect / ntotal, 2),+ results=results,+ )- return round(100 * ncorrect / ntotal, 2)def _construct_result_table(self, results: list[Tuple["dspy.Example", "dspy.Example", Any]], metric_name: strdiff --git a/dspy/teleprompt/bootstrap_finetune.py b/dspy/teleprompt/bootstrap_finetune.pyindex 117959e86..5028099f4 100644--- a/dspy/teleprompt/bootstrap_finetune.py+++ b/dspy/teleprompt/bootstrap_finetune.py@@ -231,7 +231,6 @@ def bootstrap_trace_data(devset=dataset,num_threads=num_threads,display_progress=True,- return_outputs=True,provide_traceback=False, # TODO(check with team)max_errors=len(dataset) * 10, # TODO(check with team)failure_score=failure_score,@@ -290,10 +289,10 @@ def bootstrap_trace_data(return failed_pred, trace- _, outputs = evaluator(wrapped_program, metric=wrapped_metric)+ results = evaluator(wrapped_program, metric=wrapped_metric).resultsdata = []- for example_ind, (example, prediction, score) in enumerate(outputs):+ for example_ind, (example, prediction, score) in enumerate(results):try:prediction, trace = predictionexcept ValueError as ve:diff --git a/dspy/teleprompt/copro_optimizer.py b/dspy/teleprompt/copro_optimizer.pyindex 6a3760d8d..65d09dc6f 100644--- a/dspy/teleprompt/copro_optimizer.py+++ b/dspy/teleprompt/copro_optimizer.py@@ -225,7 +225,7 @@ class COPRO(Teleprompter):f"At Depth {d+1}/{self.depth}, Evaluating Prompt Candidate #{c_i+1}/{len(candidates_)} for "f"Predictor {p_i+1} of {len(module.predictors())}.",)- score = evaluate(module_clone, devset=trainset, **eval_kwargs)+ score = evaluate(module_clone, devset=trainset, **eval_kwargs).scoreif self.prompt_model:logger.debug(f"prompt_model.inspect_history(n=1) {self.prompt_model.inspect_history(n=1)}")total_calls += 1diff --git a/dspy/teleprompt/grpo.py b/dspy/teleprompt/grpo.pyindex 4a2e70668..7745d264d 100644--- a/dspy/teleprompt/grpo.py+++ b/dspy/teleprompt/grpo.py@@ -123,8 +123,6 @@ class GRPO(FinetuneTeleprompter):devset=valset + trainset,num_threads=self.num_threads,display_progress=True,- return_all_scores=True,- return_outputs=False,provide_traceback=False, # TODO(check with team)max_errors=len(valset)*10, # TODO(check with team)failure_score=self.failure_score@@ -134,8 +132,8 @@ class GRPO(FinetuneTeleprompter):else:logger.info(f"Evaluating the student program on the validation set after training step {step_idx + 1}/{self.num_train_steps}")valset_evaluation = valset_evaluator(student, metric=self.metric)- trainset_scores = valset_evaluation[1][len(valset):]- valset_scores = valset_evaluation[1][:len(valset)]+ trainset_scores = [r[-1] for r in valset_evaluation.results[len(valset):]]+ valset_scores = [r[-1] for r in valset_evaluation.results[:len(valset)]]trainset_agg = sum(trainset_scores) / len(trainset_scores)valset_agg = sum(valset_scores) / len(valset_scores)if step_idx == -1:@@ -151,8 +149,6 @@ class GRPO(FinetuneTeleprompter):devset=valset,num_threads=self.num_threads,display_progress=True,- return_all_scores=True,- return_outputs=False,provide_traceback=False, # TODO(check with team)max_errors=len(valset)*10, # TODO(check with team)failure_score=self.failure_score@@ -163,9 +159,9 @@ class GRPO(FinetuneTeleprompter):logger.info(f"Evaluating the student program on the validation set after training step {step_idx + 1}/{self.num_train_steps}")valset_evaluation = valset_evaluator(student, metric=self.metric)if step_idx == -1:- logger.info(f"Student program validation set score before training loop: {valset_evaluation[0]}")+ logger.info(f"Student program validation set score before training loop: {valset_evaluation.score}")else:- logger.info(f"Student program validation set score after training step {step_idx + 1}/{self.num_train_steps}: {valset_evaluation[0]}")+ logger.info(f"Student program validation set score after training step {step_idx + 1}/{self.num_train_steps}: {valset_evaluation.score}")else:# No validation set provided by userif self.report_train_scores:@@ -177,8 +173,6 @@ class GRPO(FinetuneTeleprompter):devset=trainset,num_threads=self.num_threads,display_progress=True,- return_all_scores=True,- return_outputs=False,provide_traceback=False, # TODO(check with team)max_errors=len(trainset)*10, # TODO(check with team)failure_score=self.failure_score@@ -189,9 +183,9 @@ class GRPO(FinetuneTeleprompter):logger.info(f"Evaluating the student program on the validation set after training step {step_idx + 1}/{self.num_train_steps}")valset_evaluation = valset_evaluator(student, metric=self.metric)if step_idx == -1:- logger.info(f"Student program training set score before training loop: {valset_evaluation[0]}")+ logger.info(f"Student program training set score before training loop: {valset_evaluation.score}")else:- logger.info(f"Student program training set score after training step {step_idx + 1}/{self.num_train_steps}: {valset_evaluation[0]}")+ logger.info(f"Student program training set score after training step {step_idx + 1}/{self.num_train_steps}: {valset_evaluation.score}")else:# No valset provided, and not using train as valassert not self.use_train_as_val, "If report_train_scores is False, use_train_as_val must be False."diff --git a/dspy/teleprompt/infer_rules.py b/dspy/teleprompt/infer_rules.pyindex 70611962d..21ed154da 100644--- a/dspy/teleprompt/infer_rules.py+++ b/dspy/teleprompt/infer_rules.py@@ -119,9 +119,8 @@ class InferRules(BootstrapFewShot):max_errors=effective_max_errors,display_table=False,display_progress=True,- return_all_scores=True,)- score, _ = evaluate(program, metric=self.metric)+ score = evaluate(program, metric=self.metric).scorereturn scorediff --git a/dspy/teleprompt/mipro_optimizer_v2.py b/dspy/teleprompt/mipro_optimizer_v2.pyindex f4013d48c..fe1225f79 100644--- a/dspy/teleprompt/mipro_optimizer_v2.py+++ b/dspy/teleprompt/mipro_optimizer_v2.py@@ -527,9 +527,7 @@ class MIPROv2(Teleprompter):)logger.info(f"== Trial {1} / {adjusted_num_trials} - Full Evaluation of Default Program ==")- default_score, _ = eval_candidate_program(- len(valset), valset, program, evaluate, self.rng, return_all_scores=True- )+ default_score = eval_candidate_program(len(valset), valset, program, evaluate, self.rng).scorelogger.info(f"Default program score: {default_score}\n")trial_logs = {}@@ -579,7 +577,7 @@ class MIPROv2(Teleprompter):# Evaluate the candidate program (on minibatch if minibatch=True)batch_size = minibatch_size if minibatch else len(valset)- score = eval_candidate_program(batch_size, valset, candidate_program, evaluate, self.rng)+ score = eval_candidate_program(batch_size, valset, candidate_program, evaluate, self.rng).scoretotal_eval_calls += batch_size# Update best score and program@@ -816,7 +814,7 @@ class MIPROv2(Teleprompter):param_score_dict, fully_evaled_param_combos)logger.info(f"Doing full eval on next top averaging program (Avg Score: {mean_score}) from minibatch trials...")- full_eval_score = eval_candidate_program(len(valset), valset, highest_mean_program, evaluate, self.rng)+ full_eval_score = eval_candidate_program(len(valset), valset, highest_mean_program, evaluate, self.rng).scorescore_data.append({"score": full_eval_score, "program": highest_mean_program, "full_eval": True})# Log full eval as a trial so that optuna can learn from the new resultsdiff --git a/dspy/teleprompt/random_search.py b/dspy/teleprompt/random_search.pyindex ce0272ac4..e189b07bd 100644--- a/dspy/teleprompt/random_search.py+++ b/dspy/teleprompt/random_search.py@@ -123,7 +123,9 @@ class BootstrapFewShotWithRandomSearch(Teleprompter):display_progress=True,)- score, subscores = evaluate(program, return_all_scores=True)+ result = evaluate(program)++ score, subscores = result.score, [output[2] for output in result.results]all_subscores.append(subscores)diff --git a/dspy/teleprompt/teleprompt_optuna.py b/dspy/teleprompt/teleprompt_optuna.pyindex 46cc12361..31724de49 100644--- a/dspy/teleprompt/teleprompt_optuna.py+++ b/dspy/teleprompt/teleprompt_optuna.py@@ -48,9 +48,9 @@ class BootstrapFewShotWithOptuna(Teleprompter):display_table=False,display_progress=True,)- score = evaluate(program2, return_all_scores=False)+ result = evaluate(program2)trial.set_user_attr("program", program2)- return score+ return result.scoredef compile(self, student, *, teacher=None, max_demos, trainset, valset=None):import optunadiff --git a/dspy/teleprompt/utils.py b/dspy/teleprompt/utils.pyindex 31f5d19a9..4d71259e1 100644--- a/dspy/teleprompt/utils.py+++ b/dspy/teleprompt/utils.py@@ -44,31 +44,24 @@ def create_minibatch(trainset, batch_size=50, rng=None):return minibatch-def eval_candidate_program(batch_size, trainset, candidate_program, evaluate, rng=None, return_all_scores=False):+def eval_candidate_program(batch_size, trainset, candidate_program, evaluate, rng=None):"""Evaluate a candidate program on the trainset, using the specified batch size."""try:# Evaluate on the full trainsetif batch_size >= len(trainset):- return evaluate(- candidate_program,- devset=trainset,- return_all_scores=return_all_scores,- callback_metadata={"metric_key": "eval_full"},- )+ return evaluate(candidate_program, devset=trainset, callback_metadata={"metric_key": "eval_full"})# Or evaluate on a minibatchelse:return evaluate(candidate_program,devset=create_minibatch(trainset, batch_size, rng),- return_all_scores=return_all_scores,- callback_metadata={"metric_key": "eval_minibatch"},+ callback_metadata={"metric_key": "eval_minibatch"})except Exception:logger.error("An exception occurred during evaluation", exc_info=True)- if return_all_scores:- return 0.0, [0.0] * len(trainset)- return 0.0 # TODO: Handle this better, as -ve scores are possible+ # TODO: Handle this better, as -ve scores are possible+ return dspy.Prediction(score=0.0, results=[])def eval_candidate_program_with_pruning(diff --git a/tests/evaluate/test_evaluate.py b/tests/evaluate/test_evaluate.pyindex e84f25422..62572685f 100644--- a/tests/evaluate/test_evaluate.py+++ b/tests/evaluate/test_evaluate.py@@ -5,7 +5,7 @@ from unittest.mock import patchimport pytestimport dspy-from dspy.evaluate.evaluate import Evaluate+from dspy.evaluate.evaluate import Evaluate, EvaluationResultfrom dspy.evaluate.metrics import answer_exact_matchfrom dspy.predict import Predictfrom dspy.utils.callback import BaseCallback@@ -51,7 +51,7 @@ def test_evaluate_call():display_progress=False,)score = ev(program)- assert score == 100.0+ assert score.score == 100.0@pytest.mark.extra@@ -91,8 +91,8 @@ def test_multithread_evaluate_call():display_progress=False,num_threads=2,)- score = ev(program)- assert score == 100.0+ result = ev(program)+ assert result.score == 100.0def test_multi_thread_evaluate_call_cancelled(monkeypatch):@@ -129,11 +129,10 @@ def test_multi_thread_evaluate_call_cancelled(monkeypatch):display_progress=False,num_threads=2,)- score = ev(program)- assert score == 100.0+ ev(program)-def test_evaluate_call_bad():+def test_evaluate_call_wrong_answer():dspy.settings.configure(lm=DummyLM({"What is 1+1?": {"answer": "0"}, "What is 2+2?": {"answer": "0"}}))devset = [new_example("What is 1+1?", "2"), new_example("What is 2+2?", "4")]program = Predict("question -> answer")@@ -142,8 +141,8 @@ def test_evaluate_call_bad():metric=answer_exact_match,display_progress=False,)- score = ev(program)- assert score == 0.0+ result = ev(program)+ assert result.score == 0.0@pytest.mark.extra@@ -247,9 +246,13 @@ def test_evaluate_callback():metric=answer_exact_match,display_progress=False,)- score = ev(program)- assert score == 100.0+ result = ev(program)+ assert result.score == 100.0assert callback.start_call_inputs["program"] == programassert callback.start_call_count == 1- assert callback.end_call_outputs == 100.0+ assert callback.end_call_outputs.score == 100.0assert callback.end_call_count == 1++def test_evaluation_result_repr():+ result = EvaluationResult(score=100.0, results=[(new_example("What is 1+1?", "2"), {"answer": "2"}, 100.0)])+ assert repr(result) == "EvaluationResult(score=100.0, results=<list of 1 results>)"diff --git a/tests/teleprompt/test_utils.py b/tests/teleprompt/test_utils.pyindex 5b3a59098..443dc2bd8 100644--- a/tests/teleprompt/test_utils.py+++ b/tests/teleprompt/test_utils.py@@ -1,7 +1,5 @@from unittest.mock import Mock-import pytest-import dspyfrom dspy.teleprompt.utils import eval_candidate_program@@ -43,19 +41,12 @@ def test_eval_candidate_program_minibatch():assert called_kwargs["callback_metadata"] == {"metric_key": "eval_minibatch"}assert result == 0-[email protected]("return_all_scores", [True, False])-def test_eval_candidate_program_failure(return_all_scores):+def test_eval_candidate_program_failure():trainset = [1, 2, 3, 4, 5]candidate_program = DummyModule()evaluate = Mock(side_effect=ValueError("Error"))batch_size = 3- result = eval_candidate_program(- batch_size, trainset, candidate_program, evaluate, return_all_scores=return_all_scores- )+ result = eval_candidate_program(batch_size, trainset, candidate_program, evaluate)- if return_all_scores:- assert result == (0.0, [0.0] * len(trainset))- else:- assert result == 0.0+ assert result.score == 0.0