File size: 7,212 Bytes
fcdd8d2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 |
import os
# os.environ['CUDA_VISIBLE_DEVICES'] = '3'
import pathlib
from CoT.task import CoT_Task
from ToT.task import ToT_Task
from MCTS.task import MCTS_Task
import argparse
from utils.visualize import visualize
from utils.json_operator import *
from utils.verify_answer import *
from utils.self_consistency import get_consistency_output_scibench
import copy
def run(arguments):
print('-'*30, 'Begin testing', '-'*30, '\n')
# file = f'data/{arguments.task_name}/{arguments.file}.json'
# file = "/cpfs/29f69eb5e2e60f26/code/sft_intern/lh/slz/ReST-MCTS/data/math/math_500.json"
file = arguments.load_file_path
print('** file_path: ', file)
try:
data_list = load_file(file)
data_len = len(data_list)
except Exception as e:
print(f'File must be standardized json!\nError type:{e}\n')
return
assert data_len > 0, "Data list is empty!\n"
# assert 'content' in data_list[0].keys() and 'answer' in data_list[0].keys(), "Key error, Make sure json object contain correct keys!\n"
output_list = []
correct_count = 0
for i in range(data_len):
# for i in range(0, 1):
print(f'Begin to solve the problem {i+1}...\n')
# data = data_list[i]['question']
# answer = data_list[i]['answer'][0]
data = data_list[i]['question']
answer = data_list[i]['solution']
if arguments.mode == 'cot':
Task = CoT_Task(data, arguments.propose_method, arguments.value_method, arguments.temperature, evaluate=arguments.evaluate)
if arguments.consistency:
outputs = []
for cnt in range(3):
output = Task.run()
outputs.append(output)
output = get_consistency_output_scibench(outputs)
else:
output = Task.run()
elif arguments.mode == 'tot':
Task = ToT_Task(data, arguments.propose_method, arguments.value_method, arguments.algorithm,
arguments.branch, arguments.select_branch, arguments.max_depth, arguments.end_gate,
arguments.select_method, arguments.temperature, use_case_prompt=arguments.use_case_prompt,
low=arguments.low, high=arguments.high, evaluate=arguments.evaluate)
output, root = Task.run()
if arguments.visualize:
visualize(root, Task, arguments.task_name, arguments.file, i + 1)
else:
Task = MCTS_Task(data, arguments.propose_method, arguments.value_method, arguments.branch, arguments.end_gate,
arguments.roll_policy, arguments.roll_branch, arguments.roll_forward_steps, arguments.time_limit,
arguments.iteration_limit, arguments.exploration_constant, arguments.alpha, arguments.inf,
arguments.temperature, use_case_prompt=arguments.use_case_prompt, use_reflection=arguments.use_reflection,
low=arguments.low, high=arguments.high, evaluate=arguments.evaluate, answer=answer, lang='en')
output, root = Task.run()
if arguments.visualize:
visualize(root, Task, arguments.task_name, arguments.file, i + 1)
# evaluate metrics
if arguments.evaluate:
print('** output: ', output)
result = verify_float(answer, output['summary'])
output.update({'answer': answer, 'accurate': result})
if result:
print(f'The answer of problem {i+1} is correct.\n')
correct_count += 1
else:
print(f'The answer of problem {i+1} is wrong.\n')
print(f'The solution to problem {i+1} is complete.\n')
# output
base_dir = os.getcwd()
output_dir = pathlib.Path(f'{base_dir}/outputs/{arguments.task_name}/{arguments.file}/{Task.mode}')
output_file = f'{base_dir}/outputs/{arguments.task_name}/{arguments.file}/{Task.mode}/{Task.propose_method}_{Task.value_method}_{arguments.save_name}.json'
data_item = copy.deepcopy(data_list[i]) # 创建深拷贝
data_item['mcts_output'] = output
output_list.append(data_item)
pathlib.Path.mkdir(output_dir, exist_ok=True, parents=True)
dump_json(output_file, output_list)
print('** output_file: ', output_file)
print('_' * 60)
# accuracy
if args.evaluate:
print(f'Test accuracy:{correct_count / data_len}\n')
print(f'Correct number of problems:{correct_count}\nTotal number of questions:{data_len}\n')
print('_' * 60)
def parse_args():
base_args = argparse.ArgumentParser()
base_args.add_argument('--load_file_path', type=str, default='scibench')
base_args.add_argument('--task_name', type=str, default='scibench')
base_args.add_argument('--file', type=str, default='thermo_standardized') # json
base_args.add_argument('--save_name', type=str, default='test') # json
base_args.add_argument('--propose_method', type=str, choices=['gpt', 'glm', 'llama', 'local'], default='glm')
base_args.add_argument('--value_method', type=str, choices=['gpt', 'glm', 'local'], default='local')
base_args.add_argument('--mode', type=str, choices=['cot', 'tot', 'mcts'], default='tot')
base_args.add_argument('--temperature', type=float, default=0.7)
base_args.add_argument('--time_limit', type=int, default=None)
base_args.add_argument('--iteration_limit', type=int, default=100)
base_args.add_argument('--roll_policy', type=str, choices=['random', 'greedy'], default='greedy')
base_args.add_argument('--exploration_constant', type=float, default=0.4)
base_args.add_argument('--roll_forward_steps', type=int, default=2)
base_args.add_argument('--end_gate', type=float, default=0.9) # End threshold
base_args.add_argument('--branch', type=int, default=3)
base_args.add_argument('--roll_branch', type=int, default=1)
base_args.add_argument('--inf', type=float, default=0.8)
base_args.add_argument('--evaluate', type=str, default='scibench') # Whether to evaluate (empty means no evaluation)
base_args.add_argument('--alpha', type=float, default=0.5)
base_args.add_argument('--visualize', type=bool, default=False) # visualization
base_args.add_argument('--use_case_prompt', type=bool, default=False) # Use sample prompts
base_args.add_argument('--use_reflection', type=str, choices=['simple', 'common'], default='simple') # Use reflective mode
base_args.add_argument('--low', type=float, default=0)
base_args.add_argument('--high', type=float, default=1)
base_args.add_argument('--algorithm', type=str, choices=['dfs', 'bfs'], default='dfs')
base_args.add_argument('--select_branch', type=int, default=2)
base_args.add_argument('--max_depth', type=int, default=8)
base_args.add_argument('--select_method', type=str, choices=['greedy', 'sample'], default='greedy')
base_args.add_argument('--consistency', type=bool, default=True)
arguments = base_args.parse_args()
return arguments
if __name__ == '__main__':
args = parse_args()
run(args)
|