Skip to content

[Fix] Support spawn workers in LiveCodeBench evaluation #2603

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
tandede wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Fix] Support spawn workers in LiveCodeBench evaluation #2603

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f4fb053
Fix LiveCodeBench spawn worker compatibility
tandede Aug 20, 2026
3a60ccd
Merge branch 'main' into fix/livecodebench-spawn-worker
tandede Sep 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 13 additions & 13 deletions opencompass/datasets/livecodebench/evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,18 @@
DEFAULT_MEMORY_LIMIT_BYTES = 4 * 1024 * 1024 * 1024


def _run_test_in_subprocess(sample, generation, debug, result, metadata_list,
timeout, memory_limit_bytes):
from .testing_util import run_test
res, metadata = run_test(sample,
test=generation,
debug=debug,
timeout=timeout,
memory_limit_bytes=memory_limit_bytes)
result.append(res)
metadata_list.append(metadata)


def codegen_check_correctness(sample,
generation,
timeout,
Expand All @@ -33,23 +45,11 @@ def codegen_check_correctness(sample,
The global timeout is to catch some extreme/rare cases not handled by the
timeouts inside `run_test`
"""

def _temp_run(sample, generation, debug, result, metadata_list, timeout,
memory_limit_bytes):
from .testing_util import run_test
res, metadata = run_test(sample,
test=generation,
debug=debug,
timeout=timeout,
memory_limit_bytes=memory_limit_bytes)
result.append(res)
metadata_list.append(metadata)

manager = multiprocessing.Manager()
result = manager.list()
metadata_list = manager.list()
p = multiprocessing.Process(
target=_temp_run,
target=_run_test_in_subprocess,
args=(sample, generation, debug, result, metadata_list, timeout,
memory_limit_bytes),
)
Expand Down
70 changes: 59 additions & 11 deletions tests/datasets/test_livecodebench_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,12 @@
from opencompass.datasets.livecodebench import evaluator, testing_util


def _record_spawn_result(sample, generation, debug, result, metadata_list,
timeout, memory_limit_bytes):
result.append([True])
metadata_list.append({'start_method': 'spawn'})


def _run_non_linux_reliability_guard(result_queue):
import warnings

Expand Down Expand Up @@ -60,6 +66,8 @@ def test_run_test_passes_memory_limit_to_reliability_guard(self):
mock_reliability_guard.assert_called_once_with(
maximum_memory_bytes=123456)

@unittest.skipUnless('fork' in multiprocessing.get_all_start_methods(),
'requires the fork start method')
def test_codegen_check_correctness_passes_memory_limit_to_worker(self):

def fake_run_test(sample,
Expand All @@ -78,7 +86,9 @@ def fake_run_test(sample,
})
}

with patch.object(testing_util, 'run_test', fake_run_test):
fork_process = multiprocessing.get_context('fork').Process
with patch.object(testing_util, 'run_test', fake_run_test), patch.object(
evaluator.multiprocessing, 'Process', fork_process):
result, metadata = evaluator.codegen_check_correctness(
sample,
'unused generation',
Expand All @@ -89,6 +99,10 @@ def fake_run_test(sample,
self.assertEqual(result, [True])
self.assertEqual(metadata['memory_limit_bytes'], 123456)

@unittest.skipUnless('fork' in multiprocessing.get_all_start_methods(),
'requires the fork start method')
@unittest.skipIf(sys.platform == 'darwin',
'macOS rejects limits below current virtual memory')
def test_reliability_guard_adds_memory_limit_to_current_vmsize(self):
child_memory_limit = 256 * 1024 * 1024
baseline_vmsize_bytes = 64 * 1024 * 1024 * 1024
Expand Down Expand Up @@ -122,16 +136,19 @@ def fake_run_test(sample,
})
}

fork_process = multiprocessing.get_context('fork').Process
with patch.object(testing_util,
'_get_current_vmsize_bytes',
return_value=baseline_vmsize_bytes), patch.object(
testing_util, 'run_test', fake_run_test):
result, metadata = evaluator.codegen_check_correctness(
sample,
'unused generation',
timeout=1,
debug=False,
memory_limit_bytes=child_memory_limit)
return_value=baseline_vmsize_bytes):
with patch.object(testing_util, 'run_test', fake_run_test), \
patch.object(evaluator.multiprocessing, 'Process',
fork_process):
result, metadata = evaluator.codegen_check_correctness(
sample,
'unused generation',
timeout=1,
debug=False,
memory_limit_bytes=child_memory_limit)

self.assertEqual(result, [True])
effective_limit = (metadata['baseline_vmsize_bytes'] +
Expand All @@ -141,9 +158,11 @@ def fake_run_test(sample,
self.assertTrue(metadata['rlimit_data_unchanged'])
self.assertTrue(metadata['rlimit_stack_unchanged'])

@unittest.skipUnless('fork' in multiprocessing.get_all_start_methods(),
'requires the fork start method')
def test_reliability_guard_skips_memory_limit_on_non_linux(self):
result_queue = multiprocessing.Queue()
process = multiprocessing.Process(
process = multiprocessing.get_context('fork').Process(
target=_run_non_linux_reliability_guard, args=(result_queue, ))
process.start()
process.join(timeout=5)
Expand All @@ -157,6 +176,8 @@ def test_reliability_guard_skips_memory_limit_on_non_linux(self):
self.assertIn('only supported on Linux', result['warning_message'])
self.assertEqual(result['setrlimit_calls'], [])

@unittest.skipUnless('fork' in multiprocessing.get_all_start_methods(),
'requires the fork start method')
def test_codegen_check_correctness_returns_metadata_when_worker_exits(
self):

Expand All @@ -176,7 +197,9 @@ def fake_run_test(sample,
})
}

with patch.object(testing_util, 'run_test', fake_run_test):
fork_process = multiprocessing.get_context('fork').Process
with patch.object(testing_util, 'run_test', fake_run_test), patch.object(
evaluator.multiprocessing, 'Process', fork_process):
result, metadata = evaluator.codegen_check_correctness(
sample,
'unused generation',
Expand All @@ -188,6 +211,31 @@ def fake_run_test(sample,
self.assertEqual(metadata['error_message'],
'Global Timeout or Memory Limit Exceeded')

def test_codegen_check_correctness_supports_spawn_workers(self):
sample = {
'input_output':
json.dumps({
'inputs': ['1'],
'outputs': ['1'],
'fn_name': 'identity',
})
}

spawn_process = multiprocessing.get_context('spawn').Process
with patch.object(evaluator, '_run_test_in_subprocess',
_record_spawn_result), patch.object(
evaluator.multiprocessing, 'Process',
spawn_process):
result, metadata = evaluator.codegen_check_correctness(
sample,
'unused generation',
timeout=20,
debug=False,
memory_limit_bytes=None)

self.assertEqual(result, [True])
self.assertEqual(metadata['start_method'], 'spawn')


if __name__ == '__main__':
unittest.main()