# Copyright (c) Microsoft Corporation. # Licensed under the MIT license. # Mostly copied from CodeT (https://github.com/microsoft/CodeT/tree/main/CodeT) import ctypes libgcc_s = ctypes.CDLL('libgcc_s.so.1') from collections import defaultdict import multiprocessing from concurrent.futures import as_completed, ProcessPoolExecutor import logging from _execution import check_correctness, check_correctness_with_test_cases, check_spec_with_test_cases, check_with_specs_and_test_cases, check_with_specs_and_casual_inputs logging.basicConfig( format="SystemLog: [%(asctime)s][%(name)s][%(levelname)s] - %(message)s", datefmt="%Y-%m-%d %H:%M:%S", level=logging.INFO, ) logger = logging.getLogger(__name__) def evaluate_with_test_code( samples, timeout ): logger.info(f'Start evaluation with test code, timeout={timeout}') # Check the generated samples against test suites. with ProcessPoolExecutor(40) as executor: futures = [] existed_completion = defaultdict(set) results = defaultdict(defaultdict) for sample in samples: task_id = sample["task_id"] prompt = sample['prompt'] test = sample['test'] entry_point = sample['entry_point'] completion = sample["completion"] if completion in existed_completion[task_id]: continue existed_completion[task_id].add(completion) args = (task_id, prompt, completion, test, entry_point, timeout) future = executor.submit(check_correctness, *args) futures.append(future) logger.info(f'{len(futures)} execution requests are submitted') for idx, future in enumerate(as_completed(futures)): logger.info('[{}/{}] execution completed'.format(idx+1, len(futures))) result = future.result() results[result["task_id"]][result["completion"]] = result logger.info('execution finished! start parsing results') samples_with_result = [] for sample in samples: task_id = sample["task_id"] completion = sample["completion"] result = results[task_id][completion] sample["result"] = result["result"] sample["passed"] = result["passed"] samples_with_result.append(sample) assert len(samples_with_result) == len(samples), "Some problems are not attempted." return samples_with_result def evaluate_with_test_cases( solutions, test_cases_dict, timeout, assert_format=False ): logger.info(f'Start evaluation with test cases, timeout={timeout}') # Check the generated solutions against test suites. with ProcessPoolExecutor(25) as executor: futures = [] results_list = [] existed_completion = defaultdict(set) for solution in solutions: task_id = solution['task_id'] prompt = solution['prompt'] completion = solution['solution'] multi_args = solution['multi_args'] entry_point = solution['entry_point'] if completion in existed_completion[task_id]: continue existed_completion[task_id].add(completion) task_test_cases = test_cases_dict[task_id] args = (task_id, prompt, completion, task_test_cases, entry_point, timeout, multi_args, assert_format) # check_correctness_with_test_cases(*args) future = executor.submit(check_correctness_with_test_cases, *args) futures.append(future) logger.info(f'{len(futures)} execution requests are submitted') for idx, future in enumerate(as_completed(futures)): logger.info('[{}/{}] execution completed'.format(idx+1, len(futures))) result = future.result() results_list.append(result) logger.info('execution finished!') return results_list def evaluate_spec_with_test_cases( specs, test_cases_dict, timeout ): logger.info(f'Start evaluation with test cases, timeout={timeout}') # Check the generated specs against test suites. with ProcessPoolExecutor(25) as executor: futures = [] results_list = [] existed_completion = defaultdict(set) for solution in specs: task_id = solution['task_id'] prompt = solution['prompt'] spec = solution['spec'] multi_args = solution['multi_args'] if spec in existed_completion[task_id]: continue existed_completion[task_id].add(spec) task_test_cases = test_cases_dict[task_id] args = (task_id, prompt, spec, task_test_cases, timeout, multi_args) # check_spec_with_test_cases(*args) future = executor.submit(check_spec_with_test_cases, *args) futures.append(future) logger.info(f'{len(futures)} execution requests are submitted') for idx, future in enumerate(as_completed(futures)): logger.info('[{}/{}] execution completed'.format(idx+1, len(futures))) result = future.result() results_list.append(result) logger.info('execution finished!') return results_list def evaluate_with_specs_and_casual_input( solutions, specs_dict, casual_input_dict, timeout ): logger.info(f'Start evaluation with specs and casual input, timeout={timeout}') # Check the generated specs against test suites. with ProcessPoolExecutor(40) as executor: # with multiprocessing.Pool(20) as executor: futures = [] results_list = [] existed_completion = defaultdict(set) for solution in solutions: task_id = solution['task_id'] prompt = solution['prompt'] completion = solution['solution'] multi_args = solution['multi_args'] entry_point = solution['entry_point'] if completion in existed_completion[task_id]: continue existed_completion[task_id].add(completion) task_specs = specs_dict[task_id] # already been deduplicated casual_inputs = casual_input_dict[task_id] if not casual_inputs: continue for spec in task_specs: args = (task_id, prompt, completion, spec, casual_inputs, entry_point, timeout, multi_args) # check_with_specs_and_casual_inputs(*args) future = executor.submit(check_with_specs_and_casual_inputs, *args) # future = executor.apply_async(check_with_specs_and_casual_inputs, args=args) futures.append(future) logger.info(f'{len(futures)} execution requests are submitted') logger.info(f'{len(futures)} execution requests are submitted') for idx, future in enumerate(as_completed(futures)): logger.info('[{}/{}] execution completed'.format(idx+1, len(futures))) result = future.result() results_list.append(result) # for idx, future in enumerate(futures): # result = future.get() # results_list.append(result) # logger.info('[{}/{}] execution completed'.format(idx+1, len(futures))) logger.info('execution finished!') return results_list def evaluate_with_specs_and_test_input( solutions, specs_dict, test_cases_dict, timeout ): logger.info(f'Start evaluation with test cases, timeout={timeout}') # Check the generated specs against test suites. with ProcessPoolExecutor(40) as executor: futures = [] results_list = [] existed_completion = defaultdict(set) for solution in solutions: task_id = solution['task_id'] prompt = solution['prompt'] completion = solution['solution'] multi_args = solution['multi_args'] entry_point = solution['entry_point'] if completion in existed_completion[task_id]: continue existed_completion[task_id].add(completion) task_specs = specs_dict[task_id] task_test_cases = test_cases_dict[task_id] if not task_test_cases: continue for spec in task_specs: args = (task_id, prompt, completion, spec, task_test_cases, entry_point, timeout, multi_args) future = executor.submit(check_with_specs_and_test_cases, *args) futures.append(future) logger.info(f'{len(futures)} execution requests are submitted') for idx, future in enumerate(as_completed(futures)): logger.info('[{}/{}] execution completed'.format(idx+1, len(futures))) result = future.result() results_list.append(result) logger.info('execution finished!') return results_list