strands-agents-evals 0.1.1__py3-none-any.whl → 0.1.3__py3-none-any.whl

{strands_agents_evals-0.1.1.dist-info → strands_agents_evals-0.1.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: strands-agents-evals
-Version: 0.1.1
+Version: 0.1.3
 Summary: Evaluation framework for Strands
 Author-email: AWS <opensource@amazon.com>
 License: Apache-2.0

{strands_agents_evals-0.1.1.dist-info → strands_agents_evals-0.1.3.dist-info}/RECORD

@@ -1,6 +1,6 @@
 strands_evals/__init__.py,sha256=WnYsQGtkatrCKM8v_i_oCtBHNJfPaTrOg2ThUlf55Pk,485
 strands_evals/case.py,sha256=KWAL947NkmNzg9FFdTsL6KI9AFLQ8IcFjaOjcs9x5to,2131
-strands_evals/experiment.py,sha256=6gARs-JiGMSoeqC7-sjLGfL6hbEcHH5YJ4ABH0Qf3cM,28239
+strands_evals/experiment.py,sha256=yySXFW5p9xkDSvkxHBBDncxXKiuj0aDFY7iKoUyprwc,28745
 strands_evals/display/display_console.py,sha256=bOTr6RepgnifALz2DgXnnk3c4Jjxu_mA68-pFr7xry0,5932
 strands_evals/evaluators/__init__.py,sha256=OfZU5RkYewHOAnEjPKdxiEvPnfOOWNZc_9nQpAfARfI,887
 strands_evals/evaluators/evaluator.py,sha256=XEesDeT83H93B1X_w8s0Nsb1KKHy26QO8b99Hi6vKbc,7466
@@ -30,7 +30,7 @@ strands_evals/evaluators/prompt_templates/tool_selection_accuracy/tool_selection
 strands_evals/extractors/__init__.py,sha256=Jmlrk-m8sSS_LwmCVSloIkg3BjOgRzNEezjaAGMw5rw,74
 strands_evals/extractors/graph_extractor.py,sha256=TeT-58JB9roqSvy2ELz1kg8WF5YO-cfLlGZTO0F9s_4,1105
 strands_evals/extractors/swarm_extractor.py,sha256=Sm1XFCkAGVdF3XDyO3iF-20I8C6sAQ8JPNP5fgotOFU,2682
-strands_evals/extractors/tools_use_extractor.py,sha256=3WngKFdTz9XYeD0eXn90Dr1eGuM8egbOJT0w0LYxWhk,6388
+strands_evals/extractors/tools_use_extractor.py,sha256=emLL63LKldL2IA2u5wZL0ZhklZJqX0KLr5xFRt-S4i4,6600
 strands_evals/extractors/trace_extractor.py,sha256=l7gk5rUFoUcxQduPJz49OX66SdgeK1MLt81aF1yr4Lc,6653
 strands_evals/generators/__init__.py,sha256=B1F30DAIf0kPyBdE4PAZvSby-dTelqb_7hFJoATqVb0,89
 strands_evals/generators/experiment_generator.py,sha256=6wLTL0iG2b0YAiu0w8dDiaBxOIy7p_Fs7l3hCjgQc0w,22655
@@ -61,8 +61,8 @@ strands_evals/types/evaluation_report.py,sha256=vT86zO4Qn9CQbULo3aziGMdG-1qWLdcB
 strands_evals/types/trace.py,sha256=BFoEylzAlENyPH702T5MDz-_H21-Wfx-FFTSXX1tDfY,4844
 strands_evals/types/simulation/__init__.py,sha256=-mz5lW6qFfIMm4dJGaP9pXY3xeiefLbB0XevjdFykkU,133
 strands_evals/types/simulation/actor.py,sha256=ESTV8165c3Ad5QT4yYmjm-A-oZdwZ0Rf0Lq7zokjTPo,1163
-strands_agents_evals-0.1.1.dist-info/METADATA,sha256=W8UdHTxX2zsjd4F3jVuK5t2e0toxbRKBx9whF34ZjFc,17721
-strands_agents_evals-0.1.1.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-strands_agents_evals-0.1.1.dist-info/licenses/LICENSE,sha256=yIWKWnZEC7ykhOE0z330Y4XQkN0YssWOZQ-TkliALls,10141
-strands_agents_evals-0.1.1.dist-info/licenses/NOTICE,sha256=Eg13ogOmcI7JpMjxniFnKG81vwU3X8X7P_IlpvVg5RU,66
-strands_agents_evals-0.1.1.dist-info/RECORD,,
+strands_agents_evals-0.1.3.dist-info/METADATA,sha256=GnFR1FmK9no2J7NWeuV8efHNvw2eoMzl8VbSJHeRdLg,17721
+strands_agents_evals-0.1.3.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+strands_agents_evals-0.1.3.dist-info/licenses/LICENSE,sha256=yIWKWnZEC7ykhOE0z330Y4XQkN0YssWOZQ-TkliALls,10141
+strands_agents_evals-0.1.3.dist-info/licenses/NOTICE,sha256=Eg13ogOmcI7JpMjxniFnKG81vwU3X8X7P_IlpvVg5RU,66
+strands_agents_evals-0.1.3.dist-info/RECORD,,

strands_evals/experiment.py

@@ -391,8 +391,8 @@ class Experiment(Generic[InputT, OutputT]):
                     "gen_ai.evaluation.case.input": serialize(case.input),
                 },
             ) as case_span:
+                # Task execution span - execute once
                 try:
-                    # Task execution span - execute once
                     with self._tracer.start_as_current_span(
                         "task_execution",
                         attributes={
@@ -414,9 +414,21 @@ class Experiment(Generic[InputT, OutputT]):
                                 ),
                             }
                         )
-
-                    # Evaluate with each evaluator using the same task output
+                except Exception as e:
+                    case_span.record_exception(e)
                     for evaluator in self._evaluators:
+                        eval_name = evaluator.get_type_name()
+                        evaluator_data[eval_name]["cases"].append(case.model_dump())
+                        evaluator_data[eval_name]["test_passes"].append(False)
+                        evaluator_data[eval_name]["scores"].append(0)
+                        evaluator_data[eval_name]["reasons"].append(f"Task execution error: {str(e)}")
+                        evaluator_data[eval_name]["detailed_results"].append([])
+                    continue
+
+                # Evaluate with each evaluator using the same task output
+                for evaluator in self._evaluators:
+                    eval_name = evaluator.get_type_name()
+                    try:
                         with self._tracer.start_as_current_span(
                             f"evaluator {evaluator.get_type_name()}",
                             attributes={
@@ -436,21 +448,16 @@ class Experiment(Generic[InputT, OutputT]):
                                 }
                             )
 
-                            eval_name = evaluator.get_type_name()
                             evaluator_data[eval_name]["cases"].append(evaluation_context.model_dump())
                             evaluator_data[eval_name]["test_passes"].append(aggregate_pass)
                             evaluator_data[eval_name]["scores"].append(aggregate_score)
                             evaluator_data[eval_name]["reasons"].append(aggregate_reason or "")
                             evaluator_data[eval_name]["detailed_results"].append(evaluation_outputs)
-
-                except Exception as e:
-                    case_span.record_exception(e)
-                    for evaluator in self._evaluators:
-                        eval_name = evaluator.get_type_name()
-                        evaluator_data[eval_name]["cases"].append(case.model_dump())
+                    except Exception as e:
+                        evaluator_data[eval_name]["cases"].append(evaluation_context.model_dump())
                         evaluator_data[eval_name]["test_passes"].append(False)
                         evaluator_data[eval_name]["scores"].append(0)
-                        evaluator_data[eval_name]["reasons"].append(f"An error occured : {str(e)}")
+                        evaluator_data[eval_name]["reasons"].append(f"Evaluator error: {str(e)}")
                         evaluator_data[eval_name]["detailed_results"].append([])
 
         reports = []

strands_evals/extractors/tools_use_extractor.py

@@ -22,37 +22,39 @@ def extract_agent_tools_used_from_messages(agent_messages):
         if message.get("role") == "assistant":
             message_info = message.get("content")
             if len(message_info) > 0:
-                tool = None
+                tools = []
                 for message in message_info:
                     if "toolUse" in message:
+                        tools.append(message.get("toolUse"))
+
+                for tool in tools:
+                    if tool:
+                        tool_name = tool.get("name")
+                        tool_input = tool.get("input")
+                        tool_id = tool.get("toolUseId")
+                        # get the tool result from the next message
+                        tool_result = None
+                        is_error = False
+                        next_message_i = i + 1
+                        while next_message_i < len(agent_messages):
+                            next_message = agent_messages[next_message_i]
+                            next_message_i += 1
+
+                            if next_message.get("role") == "user":
+                                content = next_message.get("content")
+                                if content:
+                                    tool_result_dict = content[0].get("toolResult")
+                                    if tool_result_dict and tool_result_dict.get("toolUseId") == tool_id:
+                                        tool_result_content = tool_result_dict.get("content", [])
+                                        if len(tool_result_content) > 0:
+                                            tool_result = tool_result_content[0].get("text")
+                                            is_error = tool_result_dict.get("status") == "error"
+                                            break
+
+                        tools_used.append(
+                            {"name": tool_name, "input": tool_input, "tool_result": tool_result, "is_error": is_error}
+                        )
                         tool = message.get("toolUse")
-
-                if tool:
-                    tool_name = tool.get("name")
-                    tool_input = tool.get("input")
-                    tool_id = tool.get("toolUseId")
-                    # get the tool result from the next message
-                    tool_result = None
-                    is_error = False
-                    next_message_i = i + 1
-                    while next_message_i < len(agent_messages):
-                        next_message = agent_messages[next_message_i]
-                        next_message_i += 1
-
-                        if next_message.get("role") == "user":
-                            content = next_message.get("content")
-                            if content:
-                                tool_result_dict = content[0].get("toolResult")
-                                if tool_result_dict.get("toolUseId") == tool_id:
-                                    tool_result_content = tool_result_dict.get("content", [])
-                                    if len(tool_result_content) > 0:
-                                        tool_result = tool_result_content[0].get("text")
-                                        is_error = tool_result_dict.get("status") == "error"
-                                        break
-
-                    tools_used.append(
-                        {"name": tool_name, "input": tool_input, "tool_result": tool_result, "is_error": is_error}
-                    )
     return tools_used