run_infer.py

import asyncio
import copy
import json
import os
import tempfile
from typing import Any

import pandas as pd
import toml
from datasets import load_dataset

import openhands.agenthub
from evaluation.benchmarks.swe_bench.resource.mapping import (
    get_instance_resource_factor,
)
from evaluation.utils.shared import (
    EvalException,
    EvalMetadata,
    EvalOutput,
    assert_and_raise,
    codeact_user_response,
    get_default_sandbox_config_for_eval,
    get_metrics,
    is_fatal_evaluation_error,
    make_metadata,
    prepare_dataset,
    reset_logger_for_multiprocessing,
    run_evaluation,
    update_llm_config_for_completions_logging,
)
from openhands.controller.state.state import State
from openhands.core.config import (
    AgentConfig,
    AppConfig,
    get_llm_config_arg,
    get_parser,
)
from openhands.core.logger import openhands_logger as logger
from openhands.core.main import create_runtime, run_controller
from openhands.critic import AgentFinishedCritic
from openhands.events.action import CmdRunAction, MessageAction
from openhands.events.observation import CmdOutputObservation, ErrorObservation
from openhands.events.serialization.event import event_from_dict, event_to_dict
from openhands.runtime.base import Runtime
from openhands.utils.async_utils import call_async_from_sync
from openhands.utils.shutdown_listener import sleep_if_should_continue

USE_HINT_TEXT = os.environ.get('USE_HINT_TEXT', 'false').lower() == 'true'
RUN_WITH_BROWSING = os.environ.get('RUN_WITH_BROWSING', 'false').lower() == 'true'


AGENT_CLS_TO_FAKE_USER_RESPONSE_FN = {
    'CodeActAgent': codeact_user_response,
}


def _get_swebench_workspace_dir_name(instance: pd.Series) -> str:
    return f'{instance.repo}__{instance.version}'.replace('/', '__')


def get_instruction(instance: pd.Series, metadata: EvalMetadata) -> MessageAction:
    workspace_dir_name = _get_swebench_workspace_dir_name(instance)
    instruction = f"""
<uploaded_files>
/workspace/{workspace_dir_name}
</uploaded_files>

I've uploaded a python code repository in the directory {workspace_dir_name}. Consider the following issue description:

<issue_description>
{instance.problem_statement}
</issue_description>

Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?
I've already taken care of all changes to any of the test files described in the <issue_description>. This means you DON'T have to modify the testing logic or any of the tests in any way!
Also the development Python environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.
Your task is to make the minimal changes to non-test files in the /workspace/{workspace_dir_name} directory to ensure the <issue_description> is satisfied.

Follow these steps to resolve the issue:

1. EXPLORATION: First, thoroughly explore the repository structure using tools like `find` and `grep`.
   - Identify all files mentioned in the problem statement
   - Locate where the issue occurs in the codebase
   - Understand the surrounding context and dependencies
   - Use `grep` to search for relevant functions, classes, or error messages

2. ANALYSIS: Based on your exploration, think carefully about the problem and propose 2-5 possible approaches to fix the issue.
   - Analyze the root cause of the problem
   - Consider trade-offs between different solutions
   - Select the most promising approach and explain your reasoning

3. TEST CREATION: Before implementing any fix, create a script to reproduce and verify the issue.
   - Look at existing test files in the repository to understand the test format/structure
   - Create a minimal reproduction script that demonstrates the issue
   - Run your script to confirm the error exists

4. IMPLEMENTATION: Edit the source code to implement your chosen solution.
   - Make minimal, focused changes to fix the issue

5. VERIFICATION: Test your implementation thoroughly.
   - Run your reproduction script to verify the fix works
   - Add edge cases to your test script to ensure comprehensive coverage
   - Run existing tests related to the modified code to ensure you haven't broken anything

6. FINAL REVIEW: Carefully re-read the problem description and compare your changes with the base commit {instance["base_commit"]}.
   - Ensure you've fully addressed all requirements
   - Run any tests in the repository related to:
     * The issue you are fixing
     * The files you modified
     * The functions you changed
   - If any tests fail, revise your implementation until all tests pass

Be thorough in your exploration, testing, and reasoning. It's fine if your thinking process is lengthy - quality and completeness are more important than brevity.
"""

    if RUN_WITH_BROWSING:
        instruction += (
            '<IMPORTANT!>\n'
            'You SHOULD NEVER attempt to browse the web. '
            '</IMPORTANT!>\n'
        )

    if 'image_assets' in instance:
        assets = instance['image_assets']
        assert (
            'problem_statement' in assets
        ), 'problem_statement is required in image_assets'
        image_urls = assets['problem_statement']
        return MessageAction(content=instruction, image_urls=image_urls)
    return MessageAction(content=instruction)


# TODO: migrate all swe-bench docker to ghcr.io/openhands
DEFAULT_DOCKER_IMAGE_PREFIX = os.environ.get(
    'EVAL_DOCKER_IMAGE_PREFIX', 'docker.io/xingyaoww/'
)
logger.info(f'Default docker image prefix: {DEFAULT_DOCKER_IMAGE_PREFIX}')


def get_instance_docker_image(
    instance_id: str,
    swebench_official_image: bool = False,
) -> str:
    if swebench_official_image:
        # Official SWE-Bench image
        # swebench/sweb.eval.x86_64.django_1776_django-11333:v1
        docker_image_prefix = 'docker.io/swebench/'
        repo, name = instance_id.split('__')
        image_name = f'swebench/sweb.eval.x86_64.{repo}_1776_{name}:latest'
        logger.info(f'Using official SWE-Bench image: {image_name}')
        return image_name
    else:
        # OpenHands version of the image
        docker_image_prefix = DEFAULT_DOCKER_IMAGE_PREFIX
        image_name = 'sweb.eval.x86_64.' + instance_id
        image_name = image_name.replace(
            '__', '_s_'
        )  # to comply with docker image naming convention
        return (docker_image_prefix.rstrip('/') + '/' + image_name).lower()


def get_config(
    instance: pd.Series,
    metadata: EvalMetadata,
) -> AppConfig:
    # We use a different instance image for the each instance of swe-bench eval
    use_swebench_official_image = bool(
        ('verified' in metadata.dataset.lower() or 'lite' in metadata.dataset.lower())
        and 'swe-gym' not in metadata.dataset.lower()
    )
    base_container_image = get_instance_docker_image(
        instance['instance_id'],
        swebench_official_image=use_swebench_official_image,
    )
    logger.info(
        f'Using instance container image: {base_container_image}. '
        f'Please make sure this image exists. '
        f'Submit an issue on https://github.com/All-Hands-AI/OpenHands if you run into any issues.'
    )

    sandbox_config = get_default_sandbox_config_for_eval()
    sandbox_config.base_container_image = base_container_image
    sandbox_config.enable_auto_lint = True
    sandbox_config.use_host_network = False
    # Add platform to the sandbox config to solve issue 4401
    sandbox_config.platform = 'linux/amd64'
    sandbox_config.remote_runtime_resource_factor = get_instance_resource_factor(
        dataset_name=metadata.dataset,
        instance_id=instance['instance_id'],
    )

    config = AppConfig(
        default_agent=metadata.agent_class,
        run_as_openhands=False,
        max_iterations=metadata.max_iterations,
        runtime=os.environ.get('RUNTIME', 'docker'),
        sandbox=sandbox_config,
        # do not mount workspace
        workspace_base=None,
        workspace_mount_path=None,
    )
    config.set_llm_config(
        update_llm_config_for_completions_logging(
            metadata.llm_config, metadata.eval_output_dir, instance['instance_id']
        )
    )
    agent_config = AgentConfig(
        codeact_enable_jupyter=False,
        codeact_enable_browsing=RUN_WITH_BROWSING,
        codeact_enable_llm_editor=False,
        condenser=metadata.condenser_config,
        enable_prompt_extensions=False,
    )
    config.set_agent_config(agent_config)
    return config


def initialize_runtime(
    runtime: Runtime,
    instance: pd.Series,  # this argument is not required
):
    """Initialize the runtime for the agent.

    This function is called before the runtime is used to run the agent.
    """
    logger.info('-' * 30)
    logger.info('BEGIN Runtime Initialization Fn')
    logger.info('-' * 30)
    workspace_dir_name = _get_swebench_workspace_dir_name(instance)
    obs: CmdOutputObservation

    # Set instance id
    action = CmdRunAction(
        command=f"""echo 'export SWE_INSTANCE_ID={instance['instance_id']}' >> ~/.bashrc && echo 'export PIP_CACHE_DIR=~/.cache/pip' >> ~/.bashrc && echo "alias git='git --no-pager'" >> ~/.bashrc"""
    )
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        obs.exit_code == 0, f'Failed to export SWE_INSTANCE_ID: {str(obs)}'
    )

    action = CmdRunAction(command="""export USER=$(whoami); echo USER=${USER} """)
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(obs.exit_code == 0, f'Failed to export USER: {str(obs)}')

    # inject the init script
    script_dir = os.path.dirname(__file__)

    # inject the instance info
    action = CmdRunAction(command='mkdir -p /swe_util/eval_data/instances')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        obs.exit_code == 0,
        f'Failed to create /swe_util/eval_data/instances: {str(obs)}',
    )

    swe_instance_json_name = 'swe-bench-instance.json'
    with tempfile.TemporaryDirectory() as temp_dir:
        # Construct the full path for the desired file name within the temporary directory
        temp_file_path = os.path.join(temp_dir, swe_instance_json_name)
        # Write to the file with the desired name within the temporary directory
        with open(temp_file_path, 'w') as f:
            if not isinstance(instance, dict):
                json.dump([instance.to_dict()], f)
            else:
                json.dump([instance], f)

        # Copy the file to the desired location
        runtime.copy_to(temp_file_path, '/swe_util/eval_data/instances/')

        # inject the instance swe entry
        runtime.copy_to(
            str(os.path.join(script_dir, 'scripts/setup/instance_swe_entry.sh')),
            '/swe_util/',
        )

    action = CmdRunAction(command='cat ~/.bashrc')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(obs.exit_code == 0, f'Failed to cat ~/.bashrc: {str(obs)}')

    action = CmdRunAction(command='source ~/.bashrc')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    if isinstance(obs, ErrorObservation):
        logger.error(f'Failed to source ~/.bashrc: {str(obs)}')
    assert_and_raise(obs.exit_code == 0, f'Failed to source ~/.bashrc: {str(obs)}')

    action = CmdRunAction(command='source /swe_util/instance_swe_entry.sh')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        obs.exit_code == 0,
        f'Failed to source /swe_util/instance_swe_entry.sh: {str(obs)}',
    )

    action = CmdRunAction(command=f'cd /workspace/{workspace_dir_name}')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        obs.exit_code == 0,
        f'Failed to cd to /workspace/{workspace_dir_name}: {str(obs)}',
    )

    action = CmdRunAction(command='git reset --hard')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(obs.exit_code == 0, f'Failed to git reset --hard: {str(obs)}')

    action = CmdRunAction(
        command='for remote_name in $(git remote); do git remote remove "${remote_name}"; done'
    )
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(obs.exit_code == 0, f'Failed to remove git remotes: {str(obs)}')

    action = CmdRunAction(command='which python')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        obs.exit_code == 0 and 'testbed' in obs.content,
        f'Expected to find python interpreter from testbed, but got: {str(obs)}',
    )

    logger.info('-' * 30)
    logger.info('END Runtime Initialization Fn')
    logger.info('-' * 30)


def complete_runtime(
    runtime: Runtime,
    instance: pd.Series,  # this argument is not required, but it is used to get the workspace_dir_name
) -> dict[str, Any]:
    """Complete the runtime for the agent.

    This function is called before the runtime is used to run the agent.
    If you need to do something in the sandbox to get the correctness metric after
    the agent has run, modify this function.
    """
    logger.info('-' * 30)
    logger.info('BEGIN Runtime Completion Fn')
    logger.info('-' * 30)
    obs: CmdOutputObservation
    workspace_dir_name = _get_swebench_workspace_dir_name(instance)

    action = CmdRunAction(command=f'cd /workspace/{workspace_dir_name}')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})

    if obs.exit_code == -1:
        # The previous command is still running
        # We need to kill previous command
        logger.info('The previous command is still running, trying to kill it...')
        action = CmdRunAction(command='C-c')
        obs = runtime.run_action(action)
        logger.info(obs, extra={'msg_type': 'OBSERVATION'})

        # Then run the command again
        action = CmdRunAction(command=f'cd /workspace/{workspace_dir_name}')
        action.set_hard_timeout(600)
        logger.info(action, extra={'msg_type': 'ACTION'})
        obs = runtime.run_action(action)
        logger.info(obs, extra={'msg_type': 'OBSERVATION'})

    if obs.exit_code == -1:
        # The previous command is still running
        # We need to kill previous command
        logger.info('The previous command is still running, trying to ctrl+z it...')
        action = CmdRunAction(command='C-z')
        obs = runtime.run_action(action)
        logger.info(obs, extra={'msg_type': 'OBSERVATION'})

        # Then run the command again
        action = CmdRunAction(command=f'cd /workspace/{workspace_dir_name}')
        action.set_hard_timeout(600)
        logger.info(action, extra={'msg_type': 'ACTION'})
        obs = runtime.run_action(action)
        logger.info(obs, extra={'msg_type': 'OBSERVATION'})

    assert_and_raise(
        isinstance(obs, CmdOutputObservation) and obs.exit_code == 0,
        f'Failed to cd to /workspace/{workspace_dir_name}: {str(obs)}',
    )

    action = CmdRunAction(command='git config --global core.pager ""')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        isinstance(obs, CmdOutputObservation) and obs.exit_code == 0,
        f'Failed to git config --global core.pager "": {str(obs)}',
    )

    # First check for any git repositories in subdirectories
    action = CmdRunAction(command='find . -type d -name .git -not -path "./.git"')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        isinstance(obs, CmdOutputObservation) and obs.exit_code == 0,
        f'Failed to find git repositories: {str(obs)}',
    )

    git_dirs = [p for p in obs.content.strip().split('\n') if p]
    if git_dirs:
        # Remove all .git directories in subdirectories
        for git_dir in git_dirs:
            action = CmdRunAction(command=f'rm -rf "{git_dir}"')
            action.set_hard_timeout(600)
            logger.info(action, extra={'msg_type': 'ACTION'})
            obs = runtime.run_action(action)
            logger.info(obs, extra={'msg_type': 'OBSERVATION'})
            assert_and_raise(
                isinstance(obs, CmdOutputObservation) and obs.exit_code == 0,
                f'Failed to remove git directory {git_dir}: {str(obs)}',
            )

    # add all files
    action = CmdRunAction(command='git add -A')
    action.set_hard_timeout(600)
    logger.info(action, extra={'msg_type': 'ACTION'})
    obs = runtime.run_action(action)
    logger.info(obs, extra={'msg_type': 'OBSERVATION'})
    assert_and_raise(
        isinstance(obs, CmdOutputObservation) and obs.exit_code == 0,
        f'Failed to git add -A: {str(obs)}',
    )

    n_retries = 0
    git_patch = None
    while n_retries < 5:
        action = CmdRunAction(
            command=f'git diff --no-color --cached {instance["base_commit"]}'
        )
        action.set_hard_timeout(max(300 + 100 * n_retries, 600))
        logger.info(action, extra={'msg_type': 'ACTION'})
        obs = runtime.run_action(action)
        logger.info(obs, extra={'msg_type': 'OBSERVATION'})
        n_retries += 1
        if isinstance(obs, CmdOutputObservation):
            if obs.exit_code == 0:
                git_patch = obs.content.strip()
                break
            else:
                logger.info('Failed to get git diff, retrying...')
                sleep_if_should_continue(10)
        elif isinstance(obs, ErrorObservation):
            logger.error(f'Error occurred: {obs.content}. Retrying...')
            sleep_if_should_continue(10)
        else:
            assert_and_raise(False, f'Unexpected observation type: {str(obs)}')

    assert_and_raise(git_patch is not None, 'Failed to get git diff (None)')

    logger.info('-' * 30)
    logger.info('END Runtime Completion Fn')
    logger.info('-' * 30)
    return {'git_patch': git_patch}


def process_instance(
    instance: pd.Series,
    metadata: EvalMetadata,
    reset_logger: bool = True,
    runtime_failure_count: int = 0,
) -> EvalOutput:
    config = get_config(instance, metadata)

    # Setup the logger properly, so you can run multi-processing to parallelize the evaluation
    if reset_logger:
        log_dir = os.path.join(metadata.eval_output_dir, 'infer_logs')
        reset_logger_for_multiprocessing(logger, instance.instance_id, log_dir)
    else:
        logger.info(f'Starting evaluation for instance {instance.instance_id}.')

    # Increase resource_factor with increasing attempt_id
    if runtime_failure_count > 0:
        config.sandbox.remote_runtime_resource_factor = min(
            config.sandbox.remote_runtime_resource_factor * (2**runtime_failure_count),
            8,
        )
        logger.warning(
            f'This is the {runtime_failure_count + 1}th attempt for instance {instance.instance_id}, setting resource factor to {config.sandbox.remote_runtime_resource_factor}'
        )

    metadata = copy.deepcopy(metadata)
    metadata.details['runtime_failure_count'] = runtime_failure_count
    metadata.details['remote_runtime_resource_factor'] = (
        config.sandbox.remote_runtime_resource_factor
    )

    runtime = create_runtime(config)
    call_async_from_sync(runtime.connect)

    try:
        initialize_runtime(runtime, instance)

        message_action = get_instruction(instance, metadata)

        # Here's how you can run the agent (similar to the `main` function) and get the final task state
        state: State | None = asyncio.run(
            run_controller(
                config=config,
                initial_user_action=message_action,
                runtime=runtime,
                fake_user_response_fn=AGENT_CLS_TO_FAKE_USER_RESPONSE_FN[
                    metadata.agent_class
                ],
            )
        )

        # if fatal error, throw EvalError to trigger re-run
        if is_fatal_evaluation_error(state.last_error):
            raise EvalException('Fatal error detected: ' + state.last_error)

        # ======= THIS IS SWE-Bench specific =======
        # Get git patch
        return_val = complete_runtime(runtime, instance)
        git_patch = return_val['git_patch']
        logger.info(
            f'Got git diff for instance {instance.instance_id}:\n--------\n{git_patch}\n--------'
        )
    finally:
        runtime.close()
    # ==========================================

    # ======= Attempt to evaluate the agent's edits =======
    # we use eval_infer.sh to evaluate the agent's edits, not here
    # because the agent may alter the environment / testcases
    test_result = {
        'git_patch': git_patch,
    }

    # If you are working on some simpler benchmark that only evaluates the final model output (e.g., in a MessageAction)
    # You can simply get the LAST `MessageAction` from the returned `state.history` and parse it for evaluation.
    if state is None:
        raise ValueError('State should not be None.')

    # NOTE: this is NO LONGER the event stream, but an agent history that includes delegate agent's events
    histories = [event_to_dict(event) for event in state.history]
    metrics = get_metrics(state)

    # Save the output
    instruction = message_action.content
    if message_action.image_urls:
        instruction += (
            '\n\n<image_urls>' + '\n'.join(message_action.image_urls) + '</image_urls>'
        )
    output = EvalOutput(
        instance_id=instance.instance_id,
        instruction=instruction,
        instance=instance.to_dict(),  # SWE Bench specific
        test_result=test_result,
        metadata=metadata,
        history=histories,
        metrics=metrics,
        error=state.last_error if state and state.last_error else None,
    )
    return output


def filter_dataset(dataset: pd.DataFrame, filter_column: str) -> pd.DataFrame:
    file_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'config.toml')
    if os.path.exists(file_path):
        with open(file_path, 'r') as file:
            data = toml.load(file)
            if 'selected_ids' in data:
                selected_ids = data['selected_ids']
                logger.info(
                    f'Filtering {len(selected_ids)} tasks from "selected_ids"...'
                )
                subset = dataset[dataset[filter_column].isin(selected_ids)]
                logger.info(f'Retained {subset.shape[0]} tasks after filtering')
                return subset
    skip_ids = os.environ.get('SKIP_IDS', '').split(',')
    if len(skip_ids) > 0:
        logger.info(f'Filtering {len(skip_ids)} tasks from "SKIP_IDS"...')
        return dataset[~dataset[filter_column].isin(skip_ids)]
    return dataset


if __name__ == '__main__':
    parser = get_parser()
    parser.add_argument(
        '--dataset',
        type=str,
        default='princeton-nlp/SWE-bench',
        help='data set to evaluate on, either full-test or lite-test',
    )
    parser.add_argument(
        '--split',
        type=str,
        default='test',
        help='split to evaluate on',
    )
    args, _ = parser.parse_known_args()

    # NOTE: It is preferable to load datasets from huggingface datasets and perform post-processing
    # so we don't need to manage file uploading to OpenHands's repo
    dataset = load_dataset(args.dataset, split=args.split)
    swe_bench_tests = filter_dataset(dataset.to_pandas(), 'instance_id')
    logger.info(
        f'Loaded dataset {args.dataset} with split {args.split}: {len(swe_bench_tests)} tasks'
    )
    if 'SWE-Gym' in args.dataset:
        with open(
            os.path.join(
                os.path.dirname(os.path.abspath(__file__)),
                'split',
                'swegym_verified_instances.json',
            ),
            'r',
        ) as f:
            swegym_verified_instances = json.load(f)
            swe_bench_tests = swe_bench_tests[
                swe_bench_tests['instance_id'].isin(swegym_verified_instances)
            ]
        logger.info(
            f'{len(swe_bench_tests)} tasks left after filtering for SWE-Gym verified instances'
        )

    llm_config = None
    if args.llm_config:
        llm_config = get_llm_config_arg(args.llm_config)
        llm_config.log_completions = True
        # modify_params must be False for evaluation purpose, for reproducibility and accurancy of results
        llm_config.modify_params = False

    if llm_config is None:
        raise ValueError(f'Could not find LLM config: --llm_config {args.llm_config}')

    details = {}
    _agent_cls = openhands.agenthub.Agent.get_cls(args.agent_cls)

    dataset_descrption = (
        args.dataset.replace('/', '__') + '-' + args.split.replace('/', '__')
    )
    metadata = make_metadata(
        llm_config,
        dataset_descrption,
        args.agent_cls,
        args.max_iterations,
        args.eval_note,
        args.eval_output_dir,
        details=details,
    )

    output_file = os.path.join(metadata.eval_output_dir, 'output.jsonl')
    print(f'### OUTPUT FILE: {output_file} ###')

    # Run evaluation in iterative mode:
    # If a rollout fails to output AgentFinishAction, we will try again until it succeeds OR total 3 attempts have been made.
    ITERATIVE_EVAL_MODE = (
        os.environ.get('ITERATIVE_EVAL_MODE', 'false').lower() == 'true'
    )
    ITERATIVE_EVAL_MODE_MAX_ATTEMPTS = int(
        os.environ.get('ITERATIVE_EVAL_MODE_MAX_ATTEMPTS', '3')
    )

    if not ITERATIVE_EVAL_MODE:
        # load the dataset
        instances = prepare_dataset(swe_bench_tests, output_file, args.eval_n_limit)
        if len(instances) > 0 and not isinstance(
            instances['PASS_TO_PASS'][instances['PASS_TO_PASS'].index[0]], str
        ):
            for col in ['PASS_TO_PASS', 'FAIL_TO_PASS']:
                instances[col] = instances[col].apply(lambda x: str(x))

        run_evaluation(
            instances,
            metadata,
            output_file,
            args.eval_num_workers,
            process_instance,
            timeout_seconds=8
            * 60
            * 60,  # 8 hour PER instance should be more than enough
            max_retries=5,
        )
    else:
        critic = AgentFinishedCritic()

        def get_cur_output_file_path(attempt: int) -> str:
            return (
                f'{output_file.removesuffix(".jsonl")}.critic_attempt_{attempt}.jsonl'
            )

        eval_ids = None
        for attempt in range(1, ITERATIVE_EVAL_MODE_MAX_ATTEMPTS + 1):
            cur_output_file = get_cur_output_file_path(attempt)
            logger.info(
                f'Running evaluation with critic {critic.__class__.__name__} for attempt {attempt} of {ITERATIVE_EVAL_MODE_MAX_ATTEMPTS}.'
            )

            # For deterministic eval, we set temperature to 0.1 for (>1) attempt
            # so hopefully we get slightly different results
            if attempt > 1 and metadata.llm_config.temperature == 0:
                logger.info(
                    f'Detected temperature is 0 for (>1) attempt {attempt}. Setting temperature to 0.1...'
                )
                metadata.llm_config.temperature = 0.1

            # Load instances - at first attempt, we evaluate all instances
            # On subsequent attempts, we only evaluate the instances that failed the previous attempt determined by critic
            instances = prepare_dataset(
                swe_bench_tests, cur_output_file, args.eval_n_limit, eval_ids=eval_ids
            )
            if len(instances) > 0 and not isinstance(
                instances['PASS_TO_PASS'][instances['PASS_TO_PASS'].index[0]], str
            ):
                for col in ['PASS_TO_PASS', 'FAIL_TO_PASS']:
                    instances[col] = instances[col].apply(lambda x: str(x))

            # Run evaluation - but save them to cur_output_file
            logger.info(
                f'Evaluating {len(instances)} instances for attempt {attempt}...'
            )
            run_evaluation(
                instances,
                metadata,
                cur_output_file,
                args.eval_num_workers,
                process_instance,
                timeout_seconds=8
                * 60
                * 60,  # 8 hour PER instance should be more than enough
                max_retries=5,
            )

            # When eval is done, we update eval_ids to the instances that failed the current attempt
            instances_failed = []
            logger.info(
                f'Use critic {critic.__class__.__name__} to check {len(instances)} instances for attempt {attempt}...'
            )
            with open(cur_output_file, 'r') as f:
                for line in f:
                    instance = json.loads(line)
                    history = [event_from_dict(event) for event in instance['history']]
                    critic_result = critic.evaluate(history)
                    if not critic_result.success:
                        instances_failed.append(instance['instance_id'])
            logger.info(
                f'{len(instances_failed)} instances failed the current attempt {attempt}: {instances_failed}'
            )
            eval_ids = instances_failed

            # If no instances failed, we break
            if len(instances_failed) == 0:
                break

        # Then we should aggregate the results from all attempts into the original output file
        # and remove the intermediate files
        logger.info(
            'Aggregating results from all attempts into the original output file...'
        )
        fout = open(output_file, 'w')
        added_instance_ids = set()
        for attempt in reversed(range(1, ITERATIVE_EVAL_MODE_MAX_ATTEMPTS + 1)):
            cur_output_file = get_cur_output_file_path(attempt)
            if not os.path.exists(cur_output_file):
                logger.warning(
                    f'Intermediate output file {cur_output_file} does not exist. Skipping...'
                )
                continue

            with open(cur_output_file, 'r') as f:
                for line in f:
                    instance = json.loads(line)
                    if instance['instance_id'] not in added_instance_ids:
                        fout.write(line)
                        added_instance_ids.add(instance['instance_id'])
            logger.info(
                f'Aggregated instances from {cur_output_file}. Total instances added so far: {len(added_instance_ids)}'
            )
        fout.close()
        logger.info(
            f'Done! Total {len(added_instance_ids)} instances added to {output_file}'
        )