[Feat] add multi-swe-bench (#8174)

Co-authored-by: ByteDance User <tiger@bytedance.localdomain>

evaluation/benchmarks/multi_swe_bench/README.md

@@ -0,0 +1,65 @@
+# Multi-swe-bench Evaluation with OpenHands
+
+## LLM Setup
+
+Please follow [here](../../README.md#setup).
+
+## Dataset Preparing
+
+Please download the [**Multi-SWE-Bench** dataset](https://huggingface.co/datasets/bytedance-research/Multi-SWE-Bench).
+And change the dataset following [script](scripts/data/data_change.py).
+
+```bash
+python evaluation/benchmarks/multi_swe_bench/scripts/data/data_change.py
+```
+
+## Docker image download
+
+Please download the multi-swe-bench dokcer images from [here](https://github.com/multi-swe-bench/multi-swe-bench?tab=readme-ov-file#run-evaluation).
+
+## Generate patch
+
+Please edit the [script](infer.sh) and run it.
+
+```bash
+bash evaluation/benchmarks/multi_swe_bench/infer.sh
+```
+
+Script variable explanation:
+
+- `models`, e.g. `llm.eval_gpt4_1106_preview`, is the config group name for your
+LLM settings, as defined in your `config.toml`.
+- `git-version`, e.g. `HEAD`, is the git commit hash of the OpenHands version you would
+like to evaluate. It could also be a release tag like `0.6.2`.
+- `agent`, e.g. `CodeActAgent`, is the name of the agent for benchmarks, defaulting to `CodeActAgent`.
+- `eval_limit`, e.g. `10`, limits the evaluation to the first `eval_limit` instances. By
+default, the script evaluates the (500 issues), which will no exceed the maximum of the dataset number.
+- `max_iter`, e.g. `20`, is the maximum number of iterations for the agent to run. By
+default, it is set to 50.
+- `num_workers`, e.g. `3`, is the number of parallel workers to run the evaluation. By
+default, it is set to 1.
+- `language`, the language of your evaluating dataset.
+- `dataset`, the absolute position of the dataset jsonl.
+
+The results will be generated in evaluation/evaluation_outputs/outputs/XXX/CodeActAgent/YYY/output.jsonl, you can refer to the [example](examples/output.jsonl).
+
+## Runing evaluation
+
+First, install [multi-swe-bench](https://github.com/multi-swe-bench/multi-swe-bench).
+
+```bash
+pip install multi-swe-bench
+```
+
+Second, convert the output.jsonl to patch.jsonl with [script](scripts/eval/convert.py), you can refer to the [example](examples/patch.jsonl).
+
+```bash
+python evaluation/benchmarks/multi_swe_bench/scripts/eval/convert.py
+```
+
+Finally, evaluate with multi-swe-bench.
+The config file config.json can be refer to the [example](examples/config.json) or [github](https://github.com/multi-swe-bench/multi-swe-bench/tree/main?tab=readme-ov-file#configuration-file-example).
+
+```bash
+python -m multi_swe_bench.harness.run_evaluation --config config.json
+```

evaluation/benchmarks/multi_swe_bench/eval_infer.py

@@ -0,0 +1,456 @@
+import json
+import os
+import subprocess
+import tempfile
+import time
+from functools import partial
+
+import pandas as pd
+from swebench.harness.grading import get_eval_report
+from swebench.harness.run_evaluation import (
+    APPLY_PATCH_FAIL,
+    APPLY_PATCH_PASS,
+)
+from swebench.harness.test_spec import SWEbenchInstance, TestSpec, make_test_spec
+from swebench.harness.utils import load_swebench_dataset
+from tqdm import tqdm
+
+from evaluation.benchmarks.swe_bench.resource.mapping import (
+    get_instance_resource_factor,
+)
+from evaluation.benchmarks.swe_bench.run_infer import get_instance_docker_image
+from evaluation.utils.shared import (
+    EvalMetadata,
+    EvalOutput,
+    get_default_sandbox_config_for_eval,
+    prepare_dataset,
+    reset_logger_for_multiprocessing,
+    run_evaluation,
+)
+from openhands.core.config import (
+    AppConfig,
+    LLMConfig,
+    get_parser,
+)
+from openhands.core.logger import openhands_logger as logger
+from openhands.core.main import create_runtime
+from openhands.events.action import CmdRunAction
+from openhands.events.observation import CmdOutputObservation
+from openhands.utils.async_utils import call_async_from_sync
+
+# TODO: migrate all swe-bench docker to ghcr.io/openhands
+DOCKER_IMAGE_PREFIX = os.environ.get('EVAL_DOCKER_IMAGE_PREFIX', 'docker.io/xingyaoww/')
+logger.info(f'Using docker image prefix: {DOCKER_IMAGE_PREFIX}')
+
+
+def process_git_patch(patch):
+    if not isinstance(patch, str):
+        return ''
+
+    if not patch.strip():
+        # skip empty patches
+        return ''
+
+    patch = patch.replace('\r\n', '\n')
+    # There might be some weird characters at the beginning of the patch
+    # due to some OpenHands inference command outputs
+
+    # FOR EXAMPLE:
+    # git diff --no-color --cached 895f28f9cbed817c00ab68770433170d83132d90
+    # 0
+    # diff --git a/django/db/models/sql/.backup.query.py b/django/db/models/sql/.backup.query.py
+    # new file mode 100644
+    # index 0000000000..fc13db5948
+
+    # We "find" the first line that starts with "diff" and then we remove lines before it
+    lines = patch.split('\n')
+    for i, line in enumerate(lines):
+        if line.startswith('diff --git'):
+            patch = '\n'.join(lines[i:])
+            break
+
+    patch = patch.rstrip() + '\n'  # Make sure the last line ends with a newline
+    return patch
+
+
+def get_config(metadata: EvalMetadata, instance: pd.Series) -> AppConfig:
+    # We use a different instance image for the each instance of swe-bench eval
+    base_container_image = get_instance_docker_image(instance['instance_id'])
+    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.remote_runtime_resource_factor = get_instance_resource_factor(
+        dataset_name=metadata.dataset,
+        instance_id=instance['instance_id'],
+    )
+    config = AppConfig(
+        run_as_openhands=False,
+        runtime=os.environ.get('RUNTIME', 'docker'),
+        sandbox=sandbox_config,
+        # do not mount workspace
+        workspace_base=None,
+        workspace_mount_path=None,
+    )
+    return config
+
+
+def process_instance(
+    instance: pd.Series,
+    metadata: EvalMetadata,
+    reset_logger: bool = True,
+    log_dir: str | None = None,
+    runtime_failure_count: int = 0,
+) -> EvalOutput:
+    """
+    Evaluate agent performance on a SWE-bench problem instance.
+
+    Note that this signature differs from the expected input to `run_evaluation`. Use
+    `functools.partial` to provide optional arguments before passing to the evaluation harness.
+
+    Args:
+        log_dir (str | None, default=None): Path to directory where log files will be written. Must
+        be provided if `reset_logger` is set.
+
+    Raises:
+        AssertionError: if the `reset_logger` flag is set without a provided log directory.
+    """
+    # Setup the logger properly, so you can run multi-processing to parallelize the evaluation
+    if reset_logger:
+        assert (
+            log_dir is not None
+        ), "Can't reset logger without a provided log directory."
+        os.makedirs(log_dir, exist_ok=True)
+        reset_logger_for_multiprocessing(logger, instance.instance_id, log_dir)
+    else:
+        logger.info(f'Starting evaluation for instance {instance.instance_id}.')
+
+    config = get_config(metadata, instance)
+    instance_id = instance.instance_id
+    model_patch = instance['model_patch']
+    test_spec: TestSpec = instance['test_spec']
+    logger.info(f'Starting evaluation for instance {instance_id}.')
+
+    if 'test_result' not in instance.keys():
+        instance['test_result'] = {}
+    instance['test_result']['report'] = {
+        'empty_generation': False,
+        'resolved': False,
+        'failed_apply_patch': False,
+        'error_eval': False,
+        'test_timeout': False,
+    }
+
+    if model_patch == '':
+        instance['test_result']['report']['empty_generation'] = True
+        return EvalOutput(
+            instance_id=instance_id,
+            test_result=instance['test_result'],
+            metadata=metadata,
+        )
+
+    # 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}'
+        )
+
+    try:
+        runtime = create_runtime(config)
+        call_async_from_sync(runtime.connect)
+        # Get patch and save it to /tmp/patch.diff
+        with tempfile.TemporaryDirectory() as temp_dir:
+            # Patch file
+            patch_file_path = os.path.join(temp_dir, 'patch.diff')
+            with open(patch_file_path, 'w') as f:
+                f.write(model_patch)
+            runtime.copy_to(patch_file_path, '/tmp')
+            # Eval script
+            eval_script_path = os.path.join(temp_dir, 'eval.sh')
+            with open(eval_script_path, 'w') as f:
+                f.write(test_spec.eval_script)
+            runtime.copy_to(eval_script_path, '/tmp')
+
+        # Set +x
+        action = CmdRunAction(command='chmod +x /tmp/eval.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 obs.exit_code == 0
+
+        # Apply patch
+        exec_command = (
+            'cd /testbed && '
+            "(git apply -v /tmp/patch.diff && echo 'APPLY_PATCH_PASS' || "
+            "(echo 'Failed to apply patch with git apply, trying with patch command...' && "
+            "(patch --batch --fuzz=5 -p1 -i /tmp/patch.diff && echo 'APPLY_PATCH_PASS' || "
+            "echo 'APPLY_PATCH_FAIL')))"
+        )
+        action = CmdRunAction(command=exec_command)
+        action.set_hard_timeout(600)
+        obs = runtime.run_action(action)
+        assert isinstance(obs, CmdOutputObservation)
+        apply_patch_output = obs.content
+        assert isinstance(apply_patch_output, str)
+        instance['test_result']['apply_patch_output'] = apply_patch_output
+
+        if 'APPLY_PATCH_FAIL' in apply_patch_output:
+            logger.info(f'[{instance_id}] {APPLY_PATCH_FAIL}:\n{apply_patch_output}')
+            instance['test_result']['report']['failed_apply_patch'] = True
+
+            return EvalOutput(
+                instance_id=instance_id,
+                test_result=instance['test_result'],
+                metadata=metadata,
+            )
+        elif 'APPLY_PATCH_PASS' in apply_patch_output:
+            logger.info(f'[{instance_id}] {APPLY_PATCH_PASS}:\n{apply_patch_output}')
+
+            # Run eval script in background and save output to log file
+            log_file = '/tmp/eval_output.log'
+            action = CmdRunAction(command=f'/tmp/eval.sh > {log_file} 2>&1 & echo $!')
+            action.set_hard_timeout(300)  # Short timeout just to get the process ID
+            obs = runtime.run_action(action)
+
+            if isinstance(obs, CmdOutputObservation) and obs.exit_code == 0:
+                pid = obs.content.split()[-1].strip()
+                logger.info(
+                    f'[{instance_id}] Evaluation process started with PID: {pid}'
+                )
+
+                # Poll for completion
+                start_time = time.time()
+                timeout = 1800  # 30 minutes
+                while True:
+                    seconds_elapsed = time.time() - start_time
+                    if seconds_elapsed > timeout:
+                        logger.info(
+                            f'[{instance_id}] Evaluation timed out after {timeout} seconds'
+                        )
+                        instance['test_result']['report']['test_timeout'] = True
+                        break
+                    check_action = CmdRunAction(
+                        command=f'ps -p {pid} > /dev/null; echo $?'
+                    )
+                    check_action.set_hard_timeout(300)
+                    check_obs = runtime.run_action(check_action)
+                    if (
+                        isinstance(check_obs, CmdOutputObservation)
+                        and check_obs.content.split()[-1].strip() == '1'
+                    ):
+                        logger.info(
+                            f'[{instance_id}] Evaluation process completed after {seconds_elapsed} seconds'
+                        )
+                        break
+                    logger.info(
+                        f'[{instance_id}] [{seconds_elapsed:.0f}s] Evaluation still running, waiting...'
+                    )
+                    time.sleep(30)  # Wait for 30 seconds before checking again
+
+                # Read the log file
+                cat_action = CmdRunAction(command=f'cat {log_file}')
+                cat_action.set_hard_timeout(300)
+                cat_obs = runtime.run_action(cat_action)
+
+                # Grade answer
+                if isinstance(cat_obs, CmdOutputObservation) and cat_obs.exit_code == 0:
+                    test_output = cat_obs.content
+                    assert isinstance(test_output, str)
+                    instance['test_result']['test_output'] = test_output
+
+                    # Get report from test output
+                    logger.info(f'[{instance_id}] Grading answer...')
+                    with tempfile.TemporaryDirectory() as temp_dir:
+                        # Create a directory structure that matches the expected format
+                        # NOTE: this is a hack to make the eval report format consistent
+                        # with the original SWE-Bench eval script
+                        log_dir = os.path.join(temp_dir, 'logs', instance_id.lower())
+                        os.makedirs(log_dir, exist_ok=True)
+                        test_output_path = os.path.join(log_dir, 'test_output.txt')
+                        with open(test_output_path, 'w') as f:
+                            f.write(test_output)
+                        try:
+                            _report = get_eval_report(
+                                test_spec=test_spec,
+                                prediction={
+                                    'model_patch': model_patch,
+                                    'instance_id': instance_id,
+                                },
+                                log_path=test_output_path,
+                                include_tests_status=True,
+                            )
+                            report = _report[instance_id]
+                            logger.info(
+                                f"[{instance_id}] report: {report}\nResult for {instance_id}: resolved: {report['resolved']}"
+                            )
+                            instance['test_result']['report']['resolved'] = report[
+                                'resolved'
+                            ]
+                        except Exception as e:
+                            logger.error(
+                                f'[{instance_id}] Error when getting eval report: {e}'
+                            )
+                            instance['test_result']['report']['resolved'] = False
+                            instance['test_result']['report']['error_eval'] = True
+            else:
+                logger.info(f'[{instance_id}] Error when starting eval:\n{obs.content}')
+                instance['test_result']['report']['error_eval'] = True
+
+            return EvalOutput(
+                instance_id=instance_id,
+                test_result=instance['test_result'],
+                metadata=metadata,
+            )
+        else:
+            logger.info(
+                f'[{instance_id}] Unexpected output when applying patch:\n{apply_patch_output}'
+            )
+            raise RuntimeError(
+                instance_id,
+                f'Unexpected output when applying patch:\n{apply_patch_output}',
+                logger,
+            )
+    finally:
+        runtime.close()
+
+
+if __name__ == '__main__':
+    parser = get_parser()
+    parser.add_argument(
+        '--input-file',
+        type=str,
+        help='Path to input predictions file',
+        required=True,
+    )
+    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()
+
+    # Load SWE-Bench dataset
+    full_dataset: list[SWEbenchInstance] = load_swebench_dataset(
+        args.dataset, args.split
+    )
+    instance_id_to_instance = {
+        instance['instance_id']: instance for instance in full_dataset
+    }
+    logger.info(
+        f'Loaded dataset {args.dataset} with split {args.split} to run inference on.'
+    )
+
+    # Load predictions
+    assert args.input_file.endswith('.jsonl'), 'Input file must be a jsonl file.'
+    required_fields = ['instance_id', 'model_patch', 'test_result']
+    with open(args.input_file) as f:
+        predictions = pd.DataFrame.from_records(
+            [
+                {k: v for k, v in json.loads(line).items() if k in required_fields}
+                for line in tqdm(f, desc='Loading predictions')
+            ]
+        )
+    assert (
+        'instance_id' in predictions.columns
+    ), 'Input file must contain instance_id column.'
+
+    if 'model_patch' not in predictions.columns and (
+        'test_result' in predictions.columns
+        and 'model_patch' in predictions['test_result'].iloc[0]
+    ):
+        raise ValueError(
+            'Input file must contain model_patch column OR test_result column with model_patch field.'
+        )
+    assert len(predictions['instance_id'].unique()) == len(
+        predictions
+    ), 'instance_id column must be unique.'
+
+    if 'model_patch' not in predictions.columns:
+        predictions['model_patch'] = predictions['test_result'].apply(
+            lambda x: x.get('git_patch', '')
+        )
+    assert {'instance_id', 'model_patch'}.issubset(
+        set(predictions.columns)
+    ), 'Input file must contain instance_id and model_patch columns.'
+
+    # Process model_patch
+    predictions['model_patch'] = predictions['model_patch'].apply(process_git_patch)
+
+    # Merge predictions with dataset
+    predictions['instance'] = predictions['instance_id'].apply(
+        lambda x: instance_id_to_instance[x]
+    )
+    predictions['test_spec'] = predictions['instance'].apply(make_test_spec)
+
+    # Prepare dataset
+    output_file = args.input_file.replace('.jsonl', '.swebench_eval.jsonl')
+    instances = prepare_dataset(predictions, output_file, args.eval_n_limit)
+
+    # If possible, load the relevant metadata to avoid issues with `run_evaluation`.
+    metadata: EvalMetadata | None = None
+    metadata_filepath = os.path.join(os.path.dirname(args.input_file), 'metadata.json')
+    if os.path.exists(metadata_filepath):
+        with open(metadata_filepath, 'r') as metadata_file:
+            data = metadata_file.read()
+            metadata = EvalMetadata.model_validate_json(data)
+    else:
+        # Initialize with a dummy metadata when file doesn't exist
+        metadata = EvalMetadata(
+            agent_class='dummy_agent',  # Placeholder agent class
+            llm_config=LLMConfig(model='dummy_model'),  # Minimal LLM config
+            max_iterations=1,  # Minimal iterations
+            eval_output_dir=os.path.dirname(
+                args.input_file
+            ),  # Use input file dir as output dir
+            start_time=time.strftime('%Y-%m-%d %H:%M:%S'),  # Current time
+            git_commit=subprocess.check_output(['git', 'rev-parse', 'HEAD'])
+            .decode('utf-8')
+            .strip(),  # Current commit
+            dataset=args.dataset,  # Dataset name from args
+        )
+
+    # The evaluation harness constrains the signature of `process_instance_func` but we need to
+    # pass extra information. Build a new function object to avoid issues with multiprocessing.
+    process_instance_func = partial(
+        process_instance, log_dir=output_file.replace('.jsonl', '.logs')
+    )
+
+    run_evaluation(
+        instances,
+        metadata=metadata,
+        output_file=output_file,
+        num_workers=args.eval_num_workers,
+        process_instance_func=process_instance_func,
+    )
+
+    # Load evaluated predictions & print number of resolved predictions
+    evaluated_predictions = pd.read_json(output_file, lines=True)
+    fields = ['resolved', 'failed_apply_patch', 'error_eval', 'empty_generation']
+
+    def count_report_field(row, field):
+        return row['test_result']['report'][field]
+
+    report = {}
+    for field in fields:
+        count = evaluated_predictions.apply(
+            count_report_field, args=(field,), axis=1
+        ).sum()
+        report[field] = count
+        logger.info(
+            f'# {field}: {count} / {len(evaluated_predictions)}. ({count / len(evaluated_predictions):.2%})'
+        )

evaluation/benchmarks/multi_swe_bench/examples/config.json

@@ -0,0 +1,24 @@
+{
+    "mode": "evaluation",
+    "workdir": "./data/workdir",
+    "patch_files": [
+        "./data/patches/<your_patch_file>.jsonl"
+    ],
+    "dataset_files": [
+        "./data/patches/<to_evaluate_dataset_file>.jsonl"
+    ],
+    "force_build": false,
+    "output_dir": "./data/dataset",
+    "specifics": [],
+    "skips": [],
+    "repo_dir": "./data/repos",
+    "need_clone": false,
+    "global_env": [],
+    "clear_env": true,
+    "stop_on_error": true,
+    "max_workers": 8,
+    "max_workers_build_image": 8,
+    "max_workers_run_instance": 8,
+    "log_dir": "./data/logs",
+    "log_level": "DEBUG"
+}

evaluation/benchmarks/multi_swe_bench/examples/patch.jsonl

@@ -0,0 +1,3 @@
+{"org": "ponylang", "repo": "ponyc", "number": "4595", "fix_patch": "diff --git a/src/libponyc/ast/parser.c b/src/libponyc/ast/parser.c\nindex 9852922f..2c37d6b8 100644\n--- a/src/libponyc/ast/parser.c\n+++ b/src/libponyc/ast/parser.c\n@@ -693,6 +693,7 @@ DEF(idseqsingle);\n   AST_NODE(TK_LET);\n   TOKEN(\"variable name\", TK_ID);\n   AST_NODE(TK_NONE);  // Type\n+  SET_FLAG(AST_FLAG_IN_PARENS);\n   DONE();\n \n // idseq"}
+{"org": "ponylang", "repo": "ponyc", "number": "4593", "fix_patch": "diff --git a/packages/cli/command_parser.pony b/packages/cli/command_parser.pony\nindex a5acce8e..fa97808b 100644\n--- a/packages/cli/command_parser.pony\n+++ b/packages/cli/command_parser.pony\n@@ -100,6 +100,7 @@ class CommandParser\n             | let cs: CommandSpec box =>\n               return CommandParser._sub(cs, this).\n                 _parse_command(tokens, options, args, envsmap, opt_stop)\n+// Correctly handle parent default options\n             end\n           else\n             return SyntaxError(token, \"unknown command\")"}
+{"org": "ponylang", "repo": "ponyc", "number": "4588", "fix_patch": "diff --git a/src/libponyc/expr/match.c b/src/libponyc/expr/match.c\nindex 7d16066f..c2ec7056 100644\n--- a/src/libponyc/expr/match.c\n+++ b/src/libponyc/expr/match.c\n@@ -314,8 +314,10 @@ static ast_t* make_pattern_type(pass_opt_t* opt, ast_t* pattern)\n     case TK_DONTCAREREF:\n     case TK_MATCH_CAPTURE:\n     case TK_MATCH_DONTCARE:\n+      if (ast_id(pattern_type) == TK_ISO) pattern_type = set_cap_and_ephemeral(pattern_type, TK_TRN, TK_EPHEMERAL);\n       return pattern_type;\n \n+\n     case TK_TUPLE:\n     {\n       ast_t* pattern_child = ast_child(pattern);"}

evaluation/benchmarks/multi_swe_bench/infer.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+
+
+BASE_SCRIPT="./evaluation/benchmarks/multi_swe_bench/scripts/run_infer.sh"
+
+MODELS=("aaa" "bbb" "ccc" "ddd" "fff")
+GIT_VERSION="HEAD"
+AGENT_NAME="CodeActAgent"
+EVAL_LIMIT="500"
+MAX_ITER="50"
+NUM_WORKERS="1"
+LANGUAGE="XXX"
+DATASET="XXX"
+
+
+for MODEL in "${MODELS[@]}"; do
+    echo "=============================="
+    echo "Running benchmark for MODEL: $MODEL"
+    echo "=============================="
+
+    $BASE_SCRIPT \
+        "$MODEL" \
+        "$GIT_VERSION" \
+        "$AGENT_NAME" \
+        "$EVAL_LIMIT" \
+        "$MAX_ITER" \
+        "$NUM_WORKERS" \
+        "$DATASET" \
+        "$LANGUAGE"
+
+    echo "Completed $MODEL"
+done

evaluation/benchmarks/multi_swe_bench/resource/mapping.py

@@ -0,0 +1,39 @@
+"""Mapping instance_id to resource_factor.
+
+Different instances may have different resource requirements.
+e.g., some instances may require more memory/CPU to run inference.
+This file tracks the resource requirements of different instances.
+"""
+
+import json
+import os
+
+from openhands.core.logger import openhands_logger as logger
+
+CUR_DIR = os.path.dirname(os.path.abspath(__file__))
+DEFAULT_RUNTIME_RESOURCE_FACTOR = int(
+    os.environ.get('DEFAULT_RUNTIME_RESOURCE_FACTOR', 1)
+)
+
+# dataset to resource mapping
+_global_resource_mapping: dict[str, dict[str, float]] = {}
+
+
+def get_resource_mapping(dataset_name: str) -> dict[str, float]:
+    if dataset_name not in _global_resource_mapping:
+        file_path = os.path.join(CUR_DIR, f'{dataset_name}.json')
+        if not os.path.exists(file_path):
+            logger.warning(f'Resource mapping for {dataset_name} not found.')
+            return None
+
+        with open(file_path, 'r') as f:
+            _global_resource_mapping[dataset_name] = json.load(f)
+        logger.info(f'Loaded resource mapping for {dataset_name}')
+    return _global_resource_mapping[dataset_name]
+
+
+def get_instance_resource_factor(dataset_name: str, instance_id: str) -> int:
+    resource_mapping = get_resource_mapping(dataset_name)
+    if resource_mapping is None:
+        return DEFAULT_RUNTIME_RESOURCE_FACTOR
+    return int(resource_mapping.get(instance_id, DEFAULT_RUNTIME_RESOURCE_FACTOR))

evaluation/benchmarks/multi_swe_bench/run_infer.py

@@ -0,0 +1,853 @@
+import asyncio
+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.events.action import CmdRunAction, MessageAction, FileReadAction
+from openhands.events.observation import CmdOutputObservation, ErrorObservation
+from openhands.events.serialization.event import 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
+import pdb
+
+USE_HINT_TEXT = os.environ.get('USE_HINT_TEXT', 'false').lower() == 'true'
+USE_INSTANCE_IMAGE = os.environ.get('USE_INSTANCE_IMAGE', 'true').lower() == 'true'
+RUN_WITH_BROWSING = os.environ.get('RUN_WITH_BROWSING', 'false').lower() == 'true'
+
+# TODO: migrate all swe-bench docker to ghcr.io/openhands
+# TODO: 适应所有的语言
+DOCKER_IMAGE_PREFIX = os.environ.get('EVAL_DOCKER_IMAGE_PREFIX', '')
+LANGUAGE =os.environ.get('LANGUAGE', 'python')
+logger.info(f'Using docker image prefix: {DOCKER_IMAGE_PREFIX}')
+
+
+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):
+    workspace_dir_name = _get_swebench_workspace_dir_name(instance)
+    # Prepare instruction
+
+    # Instruction based on Anthropic's official trajectory
+    # https://github.com/eschluntz/swe-bench-experiments/tree/main/evaluation/verified/20241022_tools_claude-3-5-sonnet-updated/trajs
+    instructions = {
+        "python":(
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a python code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "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.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create a script to reproduce the error and execute it with `python <filename.py>` using the BashTool, to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "java": (
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a Java code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            "Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n"
+            "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!\n"
+            "Also the development Java environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            "Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n"
+            "Follow these steps to resolve the issue:\n"
+            "1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n"
+            '2. Create a Java class to reproduce the error and execute it by first compiling with `javac <classname>.java` and then running with `java <classname>` using the BashTool, to confirm the error\n'
+            "3. Edit the sourcecode of the repo to resolve the issue.\n"
+            "4. Rerun your reproduce script or class and confirm that the error is fixed!\n"
+            "5. Think about edgecases, add comprehensive tests for them in your reproduce class or script, and run them to make sure your fix handles these cases as well.\n"
+            f"6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance['base_commit']}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n"
+            "   - The issue you are fixing\n"
+            "   - The files you modified\n"
+            "   - The functions or classes you changed\n"
+            "   Make sure all these tests pass with your changes.\n"
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "go": (
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a Go code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "Also the development Go environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create a script or a function to reproduce the error and execute it with `go run <filename.go>` using the BashTool, to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "c": (
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a C code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "Also the development C environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create a script to reproduce the error by compiling your C code (for example, using `gcc <filename.c> -o <executable>`) and then running the executable using the BashTool, to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "cpp": (
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a C++ code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "Also the development C++ environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create or adapt a small executable (e.g., a main file or a test driver) to reproduce the issue. Build and run it (for example, by using `g++ -o reproduce reproduce.cpp && ./reproduce` via the BashTool) to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "javascript": (
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a Javascript code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "Also the development Javascript environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create a script to reproduce the error and execute it with `node <filename.js>` using the BashTool, to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "typescript":(
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a Typescript code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "Also the development Typescript environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create a script to reproduce the error and execute it with `ts-node <filename.ts>` using the BashTool, to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        ),
+        "rust":(
+            '<uploaded_files>\n'
+            f'/workspace/{workspace_dir_name}\n'
+            '</uploaded_files>\n'
+            f"I've uploaded a Rust code repository in the directory {workspace_dir_name}. Consider the following issue description:\n\n"
+            f'<issue_description>\n'
+            f'{instance.problem_statement}\n'
+            '</issue_description>\n\n'
+            'Can you help me implement the necessary changes to the repository so that the requirements specified in the <issue_description> are met?\n'
+            "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!\n"
+            "Also the development Rust environment is already set up for you (i.e., all dependencies already installed), so you don't need to install other packages.\n"
+            'Your task is to make the minimal changes to non-test files in the /workspace directory to ensure the <issue_description> is satisfied.\n'
+            'Follow these steps to resolve the issue:\n'
+            '1. As a first step, it might be a good idea to explore the repo to familiarize yourself with its structure.\n'
+            '2. Create a reproduction script (or binary) that triggers the error and execute it with `cargo run --bin <filename>` using the BashTool, to confirm the error.\n'
+            '3. Edit the sourcecode of the repo to resolve the issue.\n'
+            '4. Rerun your reproduce script and confirm that the error is fixed!\n'
+            '5. Think about edgecases, add comprehensive tests for them in your reproduce script, and run them to make sure your fix handles them as well.\n'
+            f'6. Once you are done with the initial implementation, please carefully re-read the problem description and check the difference between the current code and the base commit {instance["base_commit"]}. Do you think that the issue has been completely and comprehensively solved? Write tests to check the correctness of the solution, specifically focusing on tests that may point out any remaining problems that are not yet solved. Run all of the tests in the repo and check if any of them fail, and if they do fix the code. Repeat this process of carefully reading the problem description and current implementation, testing, and fixing any problems until you are confident that the current implementation is correct. Find and run any tests in the repo that are related to:\n'
+            '   - The issue you are fixing\n'
+            '   - The files you modified\n'
+            '   - The functions you changed\n'
+            '   Make sure all these tests pass with your changes.\n'
+            "Your thinking should be thorough and so it's fine if it's very long.\n"
+        )
+    }
+    instruction = instructions.get(LANGUAGE.lower())
+
+
+    if instruction and RUN_WITH_BROWSING:
+        instruction += (
+            '<IMPORTANT!>\n'
+            'You SHOULD NEVER attempt to browse the web. '
+            '</IMPORTANT!>\n'
+        )
+    return instruction
+
+
+
+# TODO: 适应所有的语言
+# def get_instance_docker_image(instance_id: str) -> str:
+#     image_name = 'sweb.eval.x86_64.' + instance_id
+#     if LANGUAGE == 'python':
+#         image_name = image_name.replace(
+#             '__', '_s_'
+#         )  # to comply with docker image naming convention
+#         return (DOCKER_IMAGE_PREFIX.rstrip('/') + '/' + image_name).lower()
+#     else:
+#         return image_name.lower() ##加载本地的
+def get_instance_docker_image(instance: pd.Series):
+    if LANGUAGE == 'python':
+        image_name = 'sweb.eval.x86_64.' + instance['instance_id']
+        image_name = image_name.replace(
+            '__', '_s_'
+        )  # to comply with docker image naming convention
+        return (DOCKER_IMAGE_PREFIX.rstrip('/') + '/' + image_name).lower()
+    else:
+        container_name = instance.get('repo', '').lower()
+        container_name = container_name.replace('/', '_m_')
+        instance_id = instance.get('instance_id', '')
+        tag_suffix = instance_id.split('-')[-1] if instance_id else ''
+        container_tag = f"pr-{tag_suffix}"
+        # pdb.set_trace()
+        return f"mswebench/{container_name}:{container_tag}"
+        # return "kong/insomnia:pr-8284"
+        # return "'sweb.eval.x86_64.local_insomnia"
+        # return "local_insomnia_why"
+        # return "local/kong-insomnia:pr-8117"
+
+
+
+def get_config(
+    instance: pd.Series,
+    metadata: EvalMetadata,
+) -> AppConfig:
+    SWE_BENCH_CONTAINER_IMAGE = 'ghcr.io/opendevin/eval-swe-bench:full-v1.2.1'
+    if USE_INSTANCE_IMAGE:
+        # We use a different instance image for the each instance of swe-bench eval
+        # base_container_image = get_instance_docker_image(instance['instance_id'])
+        base_container_image = get_instance_docker_image(instance)
+        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.'
+        )
+    else:
+        base_container_image = SWE_BENCH_CONTAINER_IMAGE
+        logger.info(f'Using swe-bench container image: {base_container_image}')
+
+    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(
+        enable_jupyter=False,
+        enable_browsing=RUN_WITH_BROWSING,
+        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
+
+    REPO_NAME = instance['repo'].split('/')[-1]
+    # 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 && echo 'export REPO_NAME={REPO_NAME}' >> ~/.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)}'
+    )
+    # pdb.set_trace()
+    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)}')
+
+    if USE_INSTANCE_IMAGE:
+        # 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)}',
+        )
+    else:
+        action = CmdRunAction(command='source /swe_util/swe_entry.sh')
+        action.set_hard_timeout(1800)
+        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/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)}')
+    ##TODO:这里看看需不需要判断其他语言的环境
+    # 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'})
+
+    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)}',
+    )
+
+
+    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)}',
+    )
+
+    ##删除二进制文件
+    action = CmdRunAction(
+        command=f'''
+        for file in $(git status --porcelain | grep -E "^(M| M|\\?\\?|A| A)" | cut -c4-); do
+            if [ -f "$file" ] && (file "$file" | grep -q "executable" || git check-attr binary "$file" | grep -q "binary: set"); then
+                git rm -f "$file" 2>/dev/null || rm -f "$file"
+                echo "Removed: $file"
+            fi
+        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(
+        isinstance(obs, CmdOutputObservation) and obs.exit_code == 0,
+        f'Failed to remove binary files: {str(obs)}',
+    )
+
+    # pdb.set_trace()
+
+    n_retries = 0
+    git_patch = None
+    while n_retries < 5:
+        action = CmdRunAction(
+            command=f'git diff --no-color --cached {instance["base_commit"]} > patch.diff'
+        )
+        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)}')
+
+    action = FileReadAction(
+            path='patch.diff'
+        )
+    action.set_hard_timeout(max(300 + 100 * n_retries, 600))
+    logger.info(action, extra={'msg_type': 'ACTION'})
+    obs = runtime.run_action(action)
+    git_patch = obs.content
+    # pdb.set_trace() 
+
+    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}'
+        )
+    # pdb.set_trace()
+    runtime = create_runtime(config)
+    call_async_from_sync(runtime.connect)
+
+    try:
+        initialize_runtime(runtime, instance)
+
+        instruction = 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=MessageAction(content=instruction),
+                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
+    ###remove binary diffs
+    def remove_binary_diffs(patch_text):
+        lines = patch_text.splitlines()
+        cleaned_lines = []
+        block = []
+        is_binary_block = False
+
+        for line in lines:
+            if line.startswith("diff --git "):
+                if block and not is_binary_block:
+                    cleaned_lines.extend(block)
+                block = [line]
+                is_binary_block = False
+            elif "Binary files" in line:
+                is_binary_block = True
+                block.append(line)
+            else:
+                block.append(line)
+
+        if block and not is_binary_block:
+            cleaned_lines.extend(block)
+        return "\n".join(cleaned_lines)
+    git_patch = remove_binary_diffs(git_patch)
+    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
+    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__':
+    # pdb.set_trace()
+    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)
+    # dataset = load_dataset(args.dataset)
+    dataset = load_dataset("json", data_files = args.dataset)
+    dataset = dataset[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'
+    )
+
+    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} ###')
+    instances = prepare_dataset(swe_bench_tests, output_file, args.eval_n_limit)
+
+    if len(instances) > 0 and not isinstance(
+        instances['FAIL_TO_PASS'][instances['FAIL_TO_PASS'].index[0]], str
+    ):
+        for col in ['PASS_TO_PASS', 'FAIL_TO_PASS']:
+            instances[col] = instances[col].apply(lambda x: str(x))
+    # if LANGUAGE == "java": ##TODO:适配多语言的版本
+    #     for col in ['issue_numbers', 'created_at']:
+    #         instances[col] = instances[col].apply(lambda x: str(x))
+    run_evaluation(
+        instances,
+        metadata,
+        output_file,
+        args.eval_num_workers,
+        process_instance,
+        timeout_seconds=120 * 60,  # 2 hour PER instance should be more than enough
+        max_retries=5,
+    )

evaluation/benchmarks/multi_swe_bench/scripts/data/data_change.py

@@ -0,0 +1,30 @@
+import json
+
+input_file = 'XXX.jsonl'
+output_file = 'YYY.jsonl'
+
+with open(input_file, 'r', encoding='utf-8') as fin, open(output_file, 'w', encoding='utf-8') as fout:
+    for line in fin:
+        line = line.strip()
+        if not line:
+            continue
+
+        data = json.loads(line)
+        item = data
+
+        # 提取原始数据
+        org = item.get("org", "")
+        repo = item.get("repo", "")
+        number = str(item.get("number", ""))
+
+        new_item = {}
+        new_item["repo"] = f"{org}/{repo}"
+        new_item["instance_id"] = f"{org}__{repo}-{number}"
+        new_item["problem_statement"] = item["resolved_issues"][0].get("title", "") + "\n" + item["resolved_issues"][0].get("body", "")
+        new_item["FAIL_TO_PASS"] = []
+        new_item["PASS_TO_PASS"] = []
+        new_item["base_commit"] = item['base'].get("sha","")
+        new_item["version"] = "0.1" # depends
+
+        output_data = new_item
+        fout.write(json.dumps(output_data, ensure_ascii=False) + "\n")

evaluation/benchmarks/multi_swe_bench/scripts/eval/convert.py

@@ -0,0 +1,24 @@
+import json
+import re
+
+IN_FILE = 'output.jsonl'
+OUT_FILE = 'patch.jsonl'
+
+
+def main():
+    with open(IN_FILE, 'r') as fin:
+        with open(OUT_FILE, 'w') as fout:
+            for line in fin:
+                data = json.loads(line)
+                groups = re.match(r'(.*)__(.*)-(.*)', data['instance_id'])
+                patch = {
+                    'org': groups.group(1),
+                    'repo': groups.group(2),
+                    'number': groups.group(3),
+                    'fix_patch': data['test_result']['git_patch']
+                }
+                fout.write(json.dumps(patch) + '\n')
+
+
+if __name__ == '__main__':
+    main()

evaluation/benchmarks/multi_swe_bench/scripts/run_infer.sh

@@ -0,0 +1,155 @@
+#!/bin/bash
+set -eo pipefail
+
+source "evaluation/utils/version_control.sh"
+
+MODEL_CONFIG=$1
+COMMIT_HASH=$2
+AGENT=$3
+EVAL_LIMIT=$4
+MAX_ITER=$5
+NUM_WORKERS=$6
+DATASET=$7
+# SPLIT=$8
+LANGUAGE=$8
+# N_RUNS=$10
+
+if [ -z "$NUM_WORKERS" ]; then
+  NUM_WORKERS=1
+  echo "Number of workers not specified, use default $NUM_WORKERS"
+fi
+checkout_eval_branch
+
+if [ -z "$AGENT" ]; then
+  echo "Agent not specified, use default CodeActAgent"
+  AGENT="CodeActAgent"
+fi
+
+if [ -z "$MAX_ITER" ]; then
+  echo "MAX_ITER not specified, use default 100"
+  MAX_ITER=100
+fi
+
+if [ -z "$USE_INSTANCE_IMAGE" ]; then
+  echo "USE_INSTANCE_IMAGE not specified, use default true"
+  USE_INSTANCE_IMAGE=true
+fi
+
+if [ -z "$RUN_WITH_BROWSING" ]; then
+  echo "RUN_WITH_BROWSING not specified, use default false"
+  RUN_WITH_BROWSING=false
+fi
+
+
+if [ -z "$DATASET" ]; then
+  echo "DATASET not specified, use default princeton-nlp/SWE-bench_Lite"
+  DATASET="princeton-nlp/SWE-bench_Lite"
+fi
+
+if [ -z "$LANGUAGE" ]; then
+  echo "LANUGUAGE not specified, use default python"
+  LANGUAGE="python"
+fi
+
+if [ -z "$SPLIT" ]; then
+  echo "LANUGUAGE not specified, use default python"
+  SPLIT="train"
+fi
+
+##TODO:适配多语言的版本
+# if [ -z "$SPLIT" ]; then
+#   if [ "$LANGUAGE" = "python" ]; then
+#   echo "SPLIT is test as LANUGUAGE is python"
+#     SPLIT="test"
+#   elif [ "$LANGUAGE" = "java" ]; then
+#   echo "SPLIT is java_verified as LANUGUAGE is java"
+#     SPLIT="java_verified"
+#   fi
+# fi
+
+if [ -z "$EVAL_DOCKER_IMAGE_PREFIX" ]; then
+  if [ "$LANGUAGE" = "python" ]; then
+  echo "EVAL_DOCKER_IMAGE_PREFIX is docker.io/xingyaoww/ as default as LANUGUAGE is python"
+    EVAL_DOCKER_IMAGE_PREFIX="docker.io/xingyaoww/"
+  elif [ "$LANGUAGE" = "java" ]; then
+  echo "EVAL_DOCKER_IMAGE_PREFIX is java_verified as LANUGUAGE is java"
+    EVAL_DOCKER_IMAGE_PREFIX=""
+  fi
+fi
+
+export EVAL_DOCKER_IMAGE_PREFIX=$EVAL_DOCKER_IMAGE_PREFIX
+echo "EVAL_DOCKER_IMAGE_PREFIX: $EVAL_DOCKER_IMAGE_PREFIX"
+export USE_INSTANCE_IMAGE=$USE_INSTANCE_IMAGE
+echo "USE_INSTANCE_IMAGE: $USE_INSTANCE_IMAGE"
+export RUN_WITH_BROWSING=$RUN_WITH_BROWSING
+echo "RUN_WITH_BROWSING: $RUN_WITH_BROWSING"
+export LANGUAGE=$LANGUAGE
+echo "LANGUAGE: $LANGUAGE"
+
+get_openhands_version
+
+echo "AGENT: $AGENT"
+echo "OPENHANDS_VERSION: $OPENHANDS_VERSION"
+echo "MODEL_CONFIG: $MODEL_CONFIG"
+echo "DATASET: $DATASET"
+echo "SPLIT: $SPLIT"
+
+# Default to NOT use Hint
+if [ -z "$USE_HINT_TEXT" ]; then
+  export USE_HINT_TEXT=false
+fi
+echo "USE_HINT_TEXT: $USE_HINT_TEXT"
+EVAL_NOTE="$OPENHANDS_VERSION"
+# if not using Hint, add -no-hint to the eval note
+if [ "$USE_HINT_TEXT" = false ]; then
+  EVAL_NOTE="$EVAL_NOTE-no-hint"
+fi
+
+if [ "$RUN_WITH_BROWSING" = true ]; then
+  EVAL_NOTE="$EVAL_NOTE-with-browsing"
+fi
+
+if [ -n "$EXP_NAME" ]; then
+  EVAL_NOTE="$EVAL_NOTE-$EXP_NAME"
+fi
+
+function run_eval() {
+  local eval_note=$1
+  COMMAND="poetry run python evaluation/benchmarks/multi_swe_bench/run_infer.py \
+    --agent-cls $AGENT \
+    --llm-config $MODEL_CONFIG \
+    --max-iterations $MAX_ITER \
+    --eval-num-workers $NUM_WORKERS \
+    --eval-note $eval_note \
+    --dataset $DATASET \
+    --split $SPLIT"
+
+  if [ -n "$EVAL_LIMIT" ]; then
+    echo "EVAL_LIMIT: $EVAL_LIMIT"
+    COMMAND="$COMMAND --eval-n-limit $EVAL_LIMIT"
+  fi
+
+  # Run the command
+  eval $COMMAND
+}
+
+unset SANDBOX_ENV_GITHUB_TOKEN # prevent the agent from using the github token to push
+if [ -z "$N_RUNS" ]; then
+  N_RUNS=1
+  echo "N_RUNS not specified, use default $N_RUNS"
+fi
+
+# Skip runs if the run number is in the SKIP_RUNS list
+# read from env variable SKIP_RUNS as a comma separated list of run numbers
+SKIP_RUNS=(${SKIP_RUNS//,/ })
+for i in $(seq 1 $N_RUNS); do
+  if [[ " ${SKIP_RUNS[@]} " =~ " $i " ]]; then
+    echo "Skipping run $i"
+    continue
+  fi
+  current_eval_note="$EVAL_NOTE-run_$i"
+  echo "EVAL_NOTE: $current_eval_note"
+  run_eval $current_eval_note
+done
+
+checkout_original_branch

evaluation/benchmarks/multi_swe_bench/scripts/setup/compare_patch_filename.py

@@ -0,0 +1,54 @@
+"""This script compares gold patches with OpenHands-generated patches and check whether
+OpenHands found the right (set of) files to modify.
+"""
+
+import argparse
+import json
+import re
+
+
+def extract_modified_files(patch):
+    modified_files = set()
+    file_pattern = re.compile(r'^diff --git a/(.*?) b/')
+
+    for line in patch.split('\n'):
+        match = file_pattern.match(line)
+        if match:
+            modified_files.add(match.group(1))
+
+    return modified_files
+
+
+def process_report(oh_output_file):
+    succ = 0
+    fail = 0
+    for line in open(oh_output_file):
+        line = json.loads(line)
+        instance_id = line['instance_id']
+        gold_patch = line['swe_instance']['patch']
+        generated_patch = line['git_patch']
+        gold_modified_files = extract_modified_files(gold_patch)
+        # swe-bench lite only: a gold patch always contains exactly one file
+        assert len(gold_modified_files) == 1
+        generated_modified_files = extract_modified_files(generated_patch)
+
+        # Check if all files in gold_patch are also in generated_patch
+        all_files_in_generated = gold_modified_files.issubset(generated_modified_files)
+        if all_files_in_generated:
+            succ += 1
+        else:
+            fail += 1
+            print(
+                f'{instance_id}: file mismatch, gold = {gold_modified_files}, generated = {generated_modified_files}'
+            )
+    print(
+        f'\nSUMMARY: {succ} out of {succ + fail} instances found correct files to edit, success rate = {succ / float(succ + fail)}'
+    )
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--oh_output_file', help='Path to the OH output file')
+    args = parser.parse_args()
+
+    process_report(args.oh_output_file)

evaluation/benchmarks/multi_swe_bench/scripts/setup/instance_swe_entry.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+
+source ~/.bashrc
+SWEUTIL_DIR=/swe_util
+
+# FIXME: Cannot read SWE_INSTANCE_ID from the environment variable
+# SWE_INSTANCE_ID=django__django-11099
+if [ -z "$SWE_INSTANCE_ID" ]; then
+    echo "Error: SWE_INSTANCE_ID is not set." >&2
+    exit 1
+fi
+
+if [ -z "$REPO_NAME" ]; then
+    echo "Error: REPO_NAME is not set." >&2
+    exit 1
+fi
+
+# Read the swe-bench-test-lite.json file and extract the required item based on instance_id
+item=$(jq --arg INSTANCE_ID "$SWE_INSTANCE_ID" '.[] | select(.instance_id == $INSTANCE_ID)' $SWEUTIL_DIR/eval_data/instances/swe-bench-instance.json)
+
+if [[ -z "$item" ]]; then
+  echo "No item found for the provided instance ID."
+  exit 1
+fi
+
+WORKSPACE_NAME=$(echo "$item" | jq -r '(.repo | tostring) + "__" + (.version | tostring) | gsub("/"; "__")')
+
+echo "WORKSPACE_NAME: $WORKSPACE_NAME"
+
+# Clear the workspace
+if [ -d /workspace ]; then
+    rm -rf /workspace/*
+else
+    mkdir /workspace
+fi
+# Copy repo to workspace
+if [ -d /workspace/$WORKSPACE_NAME ]; then
+    rm -rf /workspace/$WORKSPACE_NAME
+fi
+mkdir -p /workspace
+cp -r /home/$REPO_NAME /workspace/$WORKSPACE_NAME
+
+# Activate instance-specific environment
+# . /opt/miniconda3/etc/profile.d/conda.sh
+# conda activate testbed

evaluation/benchmarks/multi_swe_bench/scripts/setup/prepare_swe_utils.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+set -e
+EVAL_WORKSPACE="evaluation/benchmarks/swe_bench/eval_workspace"
+mkdir -p $EVAL_WORKSPACE
+
+# 1. Prepare REPO
+echo "==== Prepare SWE-bench repo ===="
+OH_SWE_BENCH_REPO_PATH="https://github.com/All-Hands-AI/SWE-bench.git"
+OH_SWE_BENCH_REPO_BRANCH="eval"
+git clone -b $OH_SWE_BENCH_REPO_BRANCH $OH_SWE_BENCH_REPO_PATH $EVAL_WORKSPACE/OH-SWE-bench
+
+# 2. Prepare DATA
+echo "==== Prepare SWE-bench data ===="
+EVAL_IMAGE=ghcr.io/all-hands-ai/eval-swe-bench:builder_with_conda
+EVAL_WORKSPACE=$(realpath $EVAL_WORKSPACE)
+chmod +x $EVAL_WORKSPACE/OH-SWE-bench/swebench/harness/prepare_data.sh
+if [ -d $EVAL_WORKSPACE/eval_data ]; then
+    rm -r $EVAL_WORKSPACE/eval_data
+fi
+docker run \
+    -v $EVAL_WORKSPACE:/workspace \
+    -w /workspace \
+    -u $(id -u):$(id -g) \
+    -e HF_DATASETS_CACHE="/tmp" \
+    --rm -it $EVAL_IMAGE \
+    bash -c "cd OH-SWE-bench/swebench/harness && /swe_util/miniforge3/bin/conda run -n swe-bench-eval ./prepare_data.sh && mv eval_data /workspace/"

evaluation/benchmarks/multi_swe_bench/scripts/setup/swe_entry.sh

@@ -0,0 +1,96 @@
+#!/bin/bash
+
+set -e
+
+# assert user name is `root`
+if [ "$USER" != "root" ]; then
+    echo "Error: This script is intended to be run by the 'root' user only." >&2
+    exit 1
+fi
+
+source ~/.bashrc
+
+SWEUTIL_DIR=/swe_util
+
+# Create logs directory
+LOG_DIR=/openhands/logs
+mkdir -p $LOG_DIR && chmod 777 $LOG_DIR
+
+# FIXME: Cannot read SWE_INSTANCE_ID from the environment variable
+# SWE_INSTANCE_ID=django__django-11099
+if [ -z "$SWE_INSTANCE_ID" ]; then
+    echo "Error: SWE_INSTANCE_ID is not set." >&2
+    exit 1
+fi
+
+# Read the swe-bench-test-lite.json file and extract the required item based on instance_id
+item=$(jq --arg INSTANCE_ID "$SWE_INSTANCE_ID" '.[] | select(.instance_id == $INSTANCE_ID)' $SWEUTIL_DIR/eval_data/instances/swe-bench-test-lite.json)
+
+if [[ -z "$item" ]]; then
+  echo "No item found for the provided instance ID."
+  exit 1
+fi
+
+CONDA_ENV_NAME=$(echo "$item" | jq -r '.repo + "__" + .version | gsub("/"; "__")')
+
+echo "CONDA_ENV_NAME: $CONDA_ENV_NAME"
+
+SWE_TASK_DIR=/openhands/swe_tasks
+mkdir -p $SWE_TASK_DIR
+# Dump test_patch to /workspace/test.patch
+echo "$item" | jq -r '.test_patch' > $SWE_TASK_DIR/test.patch
+# Dump patch to /workspace/gold.patch
+echo "$item" | jq -r '.patch' > $SWE_TASK_DIR/gold.patch
+# Dump the item to /workspace/instance.json except for the "test_patch" and "patch" fields
+echo "$item" | jq 'del(.test_patch, .patch)' > $SWE_TASK_DIR/instance.json
+
+# Clear the workspace
+rm -rf /workspace/*
+# Copy repo to workspace
+if [ -d /workspace/$CONDA_ENV_NAME ]; then
+    rm -rf /workspace/$CONDA_ENV_NAME
+fi
+cp -r $SWEUTIL_DIR/eval_data/testbeds/$CONDA_ENV_NAME /workspace
+
+# Reset swe-bench testbed and install the repo
+. $SWEUTIL_DIR/miniforge3/etc/profile.d/conda.sh
+conda config --set changeps1 False
+conda config --append channels conda-forge
+conda activate swe-bench-eval
+
+mkdir -p $SWE_TASK_DIR/reset_testbed_temp
+mkdir -p $SWE_TASK_DIR/reset_testbed_log_dir
+SWE_BENCH_DIR=/swe_util/OH-SWE-bench
+output=$(
+    export PYTHONPATH=$SWE_BENCH_DIR && \
+    cd $SWE_BENCH_DIR && \
+    python swebench/harness/reset_swe_env.py \
+    --swe_bench_tasks $SWEUTIL_DIR/eval_data/instances/swe-bench-test.json \
+    --temp_dir $SWE_TASK_DIR/reset_testbed_temp \
+    --testbed /workspace \
+    --conda_path $SWEUTIL_DIR/miniforge3 \
+    --instance_id $SWE_INSTANCE_ID \
+    --log_dir $SWE_TASK_DIR/reset_testbed_log_dir \
+    --timeout 900 \
+    --verbose
+)
+
+REPO_PATH=$(echo "$output" | awk -F': ' '/repo_path:/ {print $2}')
+TEST_CMD=$(echo "$output" | awk -F': ' '/test_cmd:/ {print $2}')
+echo "Repo Path: $REPO_PATH"
+echo "Test Command: $TEST_CMD"
+
+echo "export SWE_BENCH_DIR=\"$SWE_BENCH_DIR\"" >> ~/.bashrc
+echo "export REPO_PATH=\"$REPO_PATH\"" >> ~/.bashrc
+echo "export TEST_CMD=\"$TEST_CMD\"" >> ~/.bashrc
+
+if [[ "$REPO_PATH" == "None" ]]; then
+    echo "Error: Failed to retrieve repository path. Tests may not have passed or output was not as expected." >&2
+    exit 1
+fi
+
+# Activate instance-specific environment
+. $SWEUTIL_DIR/miniforge3/etc/profile.d/conda.sh
+conda activate $CONDA_ENV_NAME
+
+set +e