Merge

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/how-to/evaluation-harness.md

@@ -161,7 +161,7 @@ Pour créer un workflow d'évaluation pour votre benchmark, suivez ces étapes :
            instruction=instruction,
            test_result=evaluation_result,
            metadata=metadata,
-           history=state.history.compatibility_for_eval_history_pairs(),
+           history=compatibility_for_eval_history_pairs(state.history),
            metrics=state.metrics.get() if state.metrics else None,
            error=state.last_error if state and state.last_error else None,
        )
@@ -260,7 +260,7 @@ def codeact_user_response(state: State | None) -> str:
         # vérifier si l'agent a essayé de parler à l'utilisateur 3 fois, si oui, faire savoir à l'agent qu'il peut abandonner
         user_msgs = [
             event
-            for event in state.history.get_events()
+            for event in state.history
             if isinstance(event, MessageAction) and event.source == 'user'
         ]
         if len(user_msgs) >= 2:

docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/usage/how-to/evaluation-harness.md

@@ -158,7 +158,7 @@ OpenHands 的主要入口点在 `openhands/core/main.py` 中。以下是它工
            instruction=instruction,
            test_result=evaluation_result,
            metadata=metadata,
-           history=state.history.compatibility_for_eval_history_pairs(),
+           history=compatibility_for_eval_history_pairs(state.history),
            metrics=state.metrics.get() if state.metrics else None,
            error=state.last_error if state and state.last_error else None,
        )
@@ -257,7 +257,7 @@ def codeact_user_response(state: State | None) -> str:
         # 检查代理是否已尝试与用户对话 3 次，如果是，让代理知道它可以放弃
         user_msgs = [
             event
-            for event in state.history.get_events()
+            for event in state.history
             if isinstance(event, MessageAction) and event.source == 'user'
         ]
         if len(user_msgs) >= 2:

docs/modules/usage/how-to/evaluation-harness.md

@@ -158,7 +158,7 @@ To create an evaluation workflow for your benchmark, follow these steps:
            instruction=instruction,
            test_result=evaluation_result,
            metadata=metadata,
-           history=state.history.compatibility_for_eval_history_pairs(),
+           history=compatibility_for_eval_history_pairs(state.history),
            metrics=state.metrics.get() if state.metrics else None,
            error=state.last_error if state and state.last_error else None,
        )
@@ -257,7 +257,7 @@ def codeact_user_response(state: State | None) -> str:
         # check if the agent has tried to talk to the user 3 times, if so, let the agent know it can give up
         user_msgs = [
             event
-            for event in state.history.get_events()
+            for event in state.history
             if isinstance(event, MessageAction) and event.source == 'user'
         ]
         if len(user_msgs) >= 2:

evaluation/EDA/run_infer.py

@@ -8,6 +8,7 @@
 from evaluation.utils.shared import (
     EvalMetadata,
     EvalOutput,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -34,7 +35,7 @@ def codeact_user_response_eda(state: State) -> str:
 
     # retrieve the latest model message from history
     if state.history:
-        model_guess = state.history.get_last_agent_message()
+        model_guess = state.get_last_agent_message()
 
     assert game is not None, 'Game is not initialized.'
     msg = game.generate_user_response(model_guess)
@@ -139,7 +140,7 @@ def process_instance(
     if state is None:
         raise ValueError('State should not be None.')
 
-    final_message = state.history.get_last_agent_message()
+    final_message = state.get_last_agent_message()
 
     logger.info(f'Final message: {final_message} | Ground truth: {instance["text"]}')
     test_result = game.reward()
@@ -148,7 +149,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # Save the output
     output = EvalOutput(

evaluation/agent_bench/run_infer.py

@@ -16,6 +16,7 @@
 from evaluation.utils.shared import (
     EvalMetadata,
     EvalOutput,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -242,7 +243,7 @@ def process_instance(
         raw_ans = ''
 
         # retrieve the last agent message or thought
-        for event in state.history.get_events(reverse=True):
+        for event in reversed(state.history):
             if event.source == 'agent':
                 if isinstance(event, AgentFinishAction):
                     raw_ans = event.thought
@@ -271,7 +272,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     metrics = state.metrics.get() if state.metrics else None
 

evaluation/aider_bench/run_infer.py

@@ -15,6 +15,7 @@
 from evaluation.utils.shared import (
     EvalMetadata,
     EvalOutput,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -250,7 +251,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
     metrics = state.metrics.get() if state.metrics else None
 
     # Save the output

evaluation/biocoder/run_infer.py

@@ -13,6 +13,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -299,7 +300,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     test_result['generated'] = test_result['metadata']['1_copy_change_code']
 

evaluation/bird/run_infer.py

@@ -16,6 +16,7 @@
 from evaluation.utils.shared import (
     EvalMetadata,
     EvalOutput,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -46,7 +47,7 @@ def codeact_user_response(state: State) -> str:
         # check if the agent has tried to talk to the user 3 times, if so, let the agent know it can give up
         user_msgs = [
             event
-            for event in state.history.get_events()
+            for event in state.history
             if isinstance(event, MessageAction) and event.source == 'user'
         ]
         if len(user_msgs) > 2:
@@ -431,7 +432,7 @@ def execute_sql(db_path, sql):
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # Save the output
     output = EvalOutput(

evaluation/browsing_delegation/run_infer.py

@@ -9,6 +9,7 @@
 from evaluation.utils.shared import (
     EvalMetadata,
     EvalOutput,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -89,7 +90,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # find the last delegate action
     last_delegate_action = None

evaluation/discoverybench/run_infer.py

@@ -15,6 +15,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -173,14 +174,14 @@ def initialize_runtime(runtime: Runtime, data_files: list[str]):
 
 
 def get_last_agent_finish_action(state: State) -> AgentFinishAction:
-    for event in state.history.get_events(reverse=True):
+    for event in reversed(state.history):
         if isinstance(event, AgentFinishAction):
             return event
     return None
 
 
 def get_last_message_action(state: State) -> MessageAction:
-    for event in state.history.get_events(reverse=True):
+    for event in reversed(state.history):
         if isinstance(event, MessageAction):
             return event
     return None
@@ -307,7 +308,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # DiscoveryBench Evaluation
     eval_rec = run_eval_gold_vs_gen_NL_hypo_workflow(

evaluation/gaia/run_infer.py

@@ -12,6 +12,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -166,7 +167,7 @@ def process_instance(
 
     model_answer_raw = ''
     # get the last message or thought from the agent
-    for event in state.history.get_events(reverse=True):
+    for event in reversed(state.history):
         if event.source == 'agent':
             if isinstance(event, AgentFinishAction):
                 model_answer_raw = event.thought
@@ -203,7 +204,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # Save the output
     output = EvalOutput(

evaluation/gorilla/run_infer.py

@@ -10,6 +10,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -101,7 +102,7 @@ def process_instance(
         raise ValueError('State should not be None.')
 
     # retrieve the last message from the agent
-    model_answer_raw = state.history.get_last_agent_message()
+    model_answer_raw = state.get_last_agent_message()
 
     # attempt to parse model_answer
     ast_eval_fn = instance['ast_eval']
@@ -114,7 +115,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     output = EvalOutput(
         instance_id=instance_id,

evaluation/gpqa/run_infer.py

@@ -28,6 +28,7 @@
 from evaluation.utils.shared import (
     EvalMetadata,
     EvalOutput,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -244,7 +245,7 @@ def process_instance(
         'C': False,
         'D': False,
     }
-    for event in state.history.get_events(reverse=True):
+    for event in reversed(state.history):
         if (
             isinstance(event, AgentFinishAction)
             and event.source != 'user'
@@ -300,7 +301,7 @@ def process_instance(
         instance_id=str(instance.instance_id),
         instruction=instruction,
         metadata=metadata,
-        history=state.history.compatibility_for_eval_history_pairs(),
+        history=compatibility_for_eval_history_pairs(state.history),
         metrics=metrics,
         error=state.last_error if state and state.last_error else None,
         test_result={

evaluation/humanevalfix/run_infer.py

@@ -21,6 +21,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -255,7 +256,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # Save the output
     output = EvalOutput(

evaluation/integration_tests/run_infer.py

@@ -129,7 +129,7 @@ def process_instance(
     # # result evaluation
     # # =============================================
 
-    histories = [event_to_dict(event) for event in state.history.get_events()]
+    histories = [event_to_dict(event) for event in state.history]
     test_result: TestResult = test_class.verify_result(runtime, histories)
     metrics = state.metrics.get() if state.metrics else None
 

evaluation/logic_reasoning/run_infer.py

@@ -8,6 +8,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -225,7 +226,7 @@ def process_instance(
         raise ValueError('State should not be None.')
 
     final_message = ''
-    for event in state.history.get_events(reverse=True):
+    for event in reversed(state.history):
         if isinstance(event, AgentFinishAction):
             final_message = event.thought
             break
@@ -247,7 +248,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # Save the output
     output = EvalOutput(

evaluation/miniwob/run_infer.py

@@ -11,6 +11,7 @@
     EvalMetadata,
     EvalOutput,
     codeact_user_response,
+    compatibility_for_eval_history_pairs,
     make_metadata,
     prepare_dataset,
     reset_logger_for_multiprocessing,
@@ -182,7 +183,7 @@ def process_instance(
 
     # Instruction is the first message from the USER
     instruction = ''
-    for event in state.history.get_events():
+    for event in state.history:
         if isinstance(event, MessageAction):
             instruction = event.content
             break
@@ -194,7 +195,7 @@ def process_instance(
     # history is now available as a stream of events, rather than list of pairs of (Action, Observation)
     # for compatibility with the existing output format, we can remake the pairs here
     # remove when it becomes unnecessary
-    histories = state.history.compatibility_for_eval_history_pairs()
+    histories = compatibility_for_eval_history_pairs(state.history)
 
     # Save the output
     output = EvalOutput(