File size: 29,465 Bytes
8423f7a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 |
from propositional_logic.random_gen.evaluation_access import *
from vllm import LLM, SamplingParams
import time
import json
import pickle
import os
import time
import re
import sys
from tqdm import tqdm
import argparse
from loguru import logger
from tqdm import tqdm
import concurrent.futures
import random
import subprocess
import uuid
#python -m DFS.DFS_search_with_concurrent ~/train_scripts/train_scripts_prop_serv8/v3_2_0_66_0_8/checkpoint-1600 1 ~/leandojo_project/proplogic_serv8/train_val_test/data_45w/data_5_vars/key_directory/key_20w_quantile_0_66_0_8_out_dist_test.json ~/leandojo_project/atp_research/DFS/output/outcome_basic_key_ge_080_1000.pkl
class DFS:
def __init__(
self,
num_sampled_tactics: int,
temperature,
test_theorem_list,
max_workers,
saved_file_path,
experiment_id
) -> None:
self.max_workers = max_workers
self.test_theorem_list = test_theorem_list
self.whether_backtrack = {}
self.num_sampled_tactics = num_sampled_tactics
self.count_lean_dict = {}
self.counter_success = 0
self.counter_failed = 0
self.counter_in_process = 0
self.counter_too_long = 0
self.tactic_list_tree = {}
self.theorem_object_dict = {}
self.prompts_tactic_state_list = {}
self.root = {}
self.round_count = {}
self.parent_node_of_node = {}
self.round = 0
self.saved_file_path = saved_file_path
self.cstate_round = {} #should never be used once we want to apply tactic. should only be used at the beginning of each round
self.experiment_id = experiment_id
self.key_to_be_infered = []
self.key_not_finished = []
self.prompts_entered = []
self.counter_failed_with_error = 0
self.temperature = temperature
print(f"{len(self.test_theorem_list)} many theorem loaded")
for key in tqdm(self.test_theorem_list):
sample_eval = SingleTheoremEval(5, int(key))
self.theorem_object_dict[key] = sample_eval
init_state = self.theorem_object_dict[key].get_initial_prompt()
self.parent_node_of_node[key] = {}
self.prompts_tactic_state_list[key] = [init_state]
self.root[key] = 'open' #open, success or failed
self.tactic_list_tree[key] = {}
self.tactic_list_tree[key]["state_0:"] = None
self.parent_node_of_node[key]["state_0:"] = None
self.count_lean_dict[key] = {}
self.count_lean_dict[key]['count_lean_multiple_backtrack'] = 0
self.count_lean_dict[key]['count_lean_single_backtrack'] = 0
self.count_lean_dict[key]['count_lean_tactic_success'] = 0
self.round_count[key] = 0
self.whether_backtrack[key] = False
print('initialization done')
def get_current_state_number(self, key):
string = self.theorem_object_dict[key].get_current_state_with_label()
for line in string.split('\n'):
break
return line
def get_prev_state_number(self, key):
string = self.theorem_object_dict[key].get_prev_state_with_label()
for line in string.split('\n'):
break
return line
def revise_entered_tactic(self,entered_tactic,key):
if len(entered_tactic) != 2:
assert True==False
current_state_label = self.get_current_state_number(key)
entered_tactic[0] = current_state_label[:-1] + '_tactic_0:'
return entered_tactic
def back_track_tactic(self, key):
current_state_number = self.get_current_state_number(key)
for line in current_state_number.split('\n'):
break
match = re.search(r'\d+', line)
if match:
extracted_current_integer = int(match.group())
else:
assert False, 'no number in current state'
previous_state = self.parent_node_of_node[key][current_state_number]
previous_state_to_be_checked = self.get_prev_state_number(key)
if previous_state != previous_state_to_be_checked:
assert False, f'key is {key}, during backtrack, previous state marked and previous state by system are not the same'
for line in previous_state.split('\n'):
break
match = re.search(r'\d+', line)
if match:
extracted_previous_integer = int(match.group())
else:
assert False
return f"no solution, return to state {extracted_previous_integer} [that leads to state {extracted_current_integer}]"
def revise_output_list(self, output_text):
output_line_list = output_text.split("\n")
is_tactic = False
for idx_tactic, line in enumerate(output_line_list):
if '_tactic_' in line:
is_tactic = True
break
if is_tactic==False:
#print('output, warning: no tactic')
return ['no_tactic','no_tactic']
if "::: " in output_line_list[idx_tactic]:
output_line_list[idx_tactic] = output_line_list[idx_tactic][4:]
entered_tactic_list = output_line_list[idx_tactic:idx_tactic+2]
if len(entered_tactic_list) == 1:
return ['no_tactic','no_tactic']
return entered_tactic_list
def check_if_failure_per_key(self, key):
if len(self.tactic_list_tree[key]['state_0:']) == 0 and self.get_current_state_number(key) == 'state_0:':
print('triggered failure')
return True
else:
return False
def check_path_length(self,key):
current_state_label = self.get_current_state_number(key)
previous_state_label = current_state_label
theorem_object_length = 1
#print(f'key is {key}, current tactic list tree is {self.tactic_list_tree[key]}')
while True:
#print("state chain label is: ", previous_state_label)
if previous_state_label == 'state_0:':
break
previous_state_label = self.parent_node_of_node[key][current_state_label]
theorem_object_length += 1
current_state_label = previous_state_label
#print(f'prompt tactic list length is {len(self.prompts_tactic_state_list[key])}')
#print(f'theorem_object_length is {theorem_object_length}')
if theorem_object_length != len(self.prompts_tactic_state_list[key]):
assert True==False, "path_length not equal to each other"
def check_if_program_finished(self):
stop_signal = True
for key in self.test_theorem_list:
if self.root[key] == 'open':
stop_signal = False
else:
pass
return stop_signal
def revise_prompt(self, prompts_tactic_state_list_per_key):
pattern = r'state_\d+:'
matches = re.findall(pattern, prompts_tactic_state_list_per_key)
state_order = {}
order = 0
for match in matches:
if match not in state_order:
state_order[match] = order
order += 1
for state, ord in state_order.items():
prompts_tactic_state_list_per_key = prompts_tactic_state_list_per_key.replace(state, f'state_{ord}:')
last_state_id = None
output_prompt = []
for line in prompts_tactic_state_list_per_key.split('\n'):
if re.search('state_\d+:', line):
last_state_id = line[6:-1]
elif re.search('state_\d+_tactic_', line):
line = f'state_{last_state_id}_tactic_0:'
output_prompt.append(line)
'''for idx, item in enumerate(prompts_tactic_state_list_per_key):
temp_string = re.sub(r'state_(\d+):',f'state_{idx}:', item)
prompts_tactic_state_list_per_key[idx] = re.sub(r'state_(\d+)_tactic_(\d+):',f'state_{idx-1}_tactic_0:', temp_string)'''
return "\n".join(output_prompt)
def status_report(self):
counter_in_process = 0
counter_success = 0
counter_failed = 0
counter_too_long = 0
counter_failed_with_error = 0
for key in self.test_theorem_list:
if self.root[key] == 'open':
counter_in_process += 1
if self.root[key] == 'success':
counter_success += 1
if self.root[key] == 'failed':
counter_failed += 1
if self.root[key] == 'failed, too long':
counter_too_long += 1
if self.root[key] == 'failed with error':
counter_failed_with_error += 1
self.counter_success = counter_success
self.counter_failed = counter_failed
self.counter_failed_with_error = counter_failed_with_error
self.counter_too_long = counter_too_long
self.counter_in_process = counter_in_process
if counter_success + counter_failed + counter_too_long + counter_in_process + counter_failed_with_error != len(test_theorem_list):
assert False, 'success, failed, too long, in process, failed with error add up not equal to total number'
print(f'saved_file_path is {self.saved_file_path}')
print(f'total number of theorem is {len(self.test_theorem_list)}')
print(f'proof success number is {self.counter_success}')
print(f'proof failed number is {self.counter_failed}')
print(f'proof failed with error number is {self.counter_failed_with_error}')
print(f'proof too long number is {self.counter_too_long}')
print(f'proof in process number is {self.counter_in_process}')
count_lean_single_backtrack = 0
count_lean_multiple_backtrack = 0
count_lean_tactic_success = 0
for key in test_theorem_list:
count_lean_single_backtrack += self.count_lean_dict[key]['count_lean_single_backtrack']
count_lean_multiple_backtrack += self.count_lean_dict[key]['count_lean_multiple_backtrack']
count_lean_tactic_success += self.count_lean_dict[key]['count_lean_tactic_success']
print(f'total lean count tactic success is {count_lean_tactic_success}')
print(f'total lean count single backtrack is {count_lean_single_backtrack}')
print(f'total lean count multiple backtrack is {count_lean_multiple_backtrack}')
def collect_inference_result(self, key_to_be_infered, outputs):
for idx, output_list in tqdm(enumerate(outputs), total=len(outputs), desc=f"Processing LLM output for Round {self.round}"):
assinged_output_list_per_key = []
for i in range(0, self.num_sampled_tactics):
output_tactic = self.revise_output_list(output_list.outputs[i].text)
if output_tactic[0] == 'no_tactic' or output_tactic[1] == 'no_tactic':
pass
else:
assinged_output_list_per_key.append(output_tactic)
#print(f'key is {key_to_be_infered[idx]}, output_tactic is {output_tactic}')
#print()
seen = set()
unique_assigned_output_list_per_key = []
for inner_list in assinged_output_list_per_key:
inner_tuple = tuple(inner_list)
if inner_tuple not in seen:
seen.add(inner_tuple)
unique_assigned_output_list_per_key.append(inner_list)
#print(
# f'key is {key_to_be_infered[idx]}, state_number to be assigned new tactic list is {self.cstate_round[key_to_be_infered[idx]]}')
#print(f'key is {key_to_be_infered[idx]}, Assigned tactic list is {unique_assigned_output_list_per_key}')
self.tactic_list_tree[key_to_be_infered[idx]][self.cstate_round[key_to_be_infered[idx]]] = unique_assigned_output_list_per_key
def current_state_obtained_list(self, key):
if self.root[key] == 'open':
self.cstate_round[key] = self.get_current_state_number(key)
cstate = self.cstate_round[key]
pickle.dump(cstate, open(f'~/leandojo_project/atp_research/DFS/temp/current_state_{key}_{self.round}_{self.experiment_id}.pkl','wb'))
def search(self):
tokenizer = llm.get_tokenizer()
while True:
self.round += 1
print(f'Round {self.round}------')
if self.check_if_program_finished() or self.round > 65:
print('confirmed test theorem finished. exit.')
self.status_report()
break
self.key_to_be_infered = []
self.key_not_finished = []
self.prompts_entered = []
with concurrent.futures.ProcessPoolExecutor(max_workers=self.max_workers) as executor:
list(tqdm(executor.map(self.current_state_obtained_list, self.test_theorem_list), total=len(self.test_theorem_list), desc=f"get current state for Round {self.round}"))
key_prompt_length_list = []
for key in tqdm(self.test_theorem_list, total=len(self.test_theorem_list), desc=f"check state for theorems for round {self.round}" ):
if self.root[key] == 'open':
self.round_count[key] = self.round
self.key_not_finished.append(key)
self.cstate_round[key] = pickle.load(open(f'~/leandojo_project/atp_research/DFS/temp/current_state_{key}_{self.round}_{self.experiment_id}.pkl','rb'))
if self.tactic_list_tree[key][self.cstate_round[key]] == None:
prompt_per_key = self.revise_prompt('\n'.join(self.prompts_tactic_state_list[key]))
key_prompt_length_list.append(len(prompt_per_key.split()))
tokenized_prompt_per_key = tokenizer.encode(prompt_per_key)
if len(prompt_per_key.split()) < 1500 and len(tokenized_prompt_per_key) < 4000: # used to be 1500
self.key_to_be_infered.append(key)
self.prompts_entered.append(prompt_per_key)
else:
self.root[key] = 'failed, too long'
self.key_not_finished.remove(key)
print(f'key open need inference before check length, length list is {key_prompt_length_list}')
print(f'key to be infered is {self.key_to_be_infered}')
sampling_params = SamplingParams(n=self.num_sampled_tactics, temperature=self.temperature, top_p=1,
max_tokens=200) # temperature is 1.2 at beginning
outputs = llm.generate(self.prompts_entered, sampling_params)
print('now we collect the inference')
self.collect_inference_result(self.key_to_be_infered, outputs)
print('inference collected')
print(f'enter concurrent process with max_workers as {self.max_workers}')
with concurrent.futures.ProcessPoolExecutor(max_workers=self.max_workers) as executor:
list(tqdm(executor.map(self.search_per_key_per_step, self.key_not_finished), total=len(self.key_not_finished), desc=f"Lean_verifying for Round {self.round}"))
'''for key in self.key_not_finished:
self.search_per_key_per_step(key)'''
for key in tqdm(self.key_not_finished, total=len(self.key_not_finished), desc=f"Retrieve results from concurrent processing for Round {self.round}"):
report_key_file = pickle.load(open(f'~/leandojo_project/atp_research/DFS/temp/temp_file_{key}_{self.round}_{self.experiment_id}.pkl','rb'))
self.count_lean_dict[key]['count_lean_single_backtrack'] += report_key_file['count_lean_single_backtrack']
self.count_lean_dict[key]['count_lean_multiple_backtrack'] += report_key_file['count_lean_multiple_backtrack']
self.count_lean_dict[key]['count_lean_tactic_success'] += report_key_file['count_lean_tactic_success']
self.root[key] = report_key_file['key_status']
self.tactic_list_tree[key] = report_key_file['tactic_list_tree']
self.prompts_tactic_state_list[key] = report_key_file['prompts_tactic_state_list']
self.theorem_object_dict[key] = report_key_file['theorem_object_dict']
self.parent_node_of_node[key] = report_key_file['node_relation']
self.whether_backtrack[key] = report_key_file['whether_backtrack']
self.status_report()
self.save_outcome()
def search_per_key_per_step(self, key):
#print(f'key is {key}, current tactic_list_tree for Round {self.round} is {self.tactic_list_tree[key]}')
try:
key_status = 'open'
count_lean_tactic = 0
count_lean_single_backtrack = 0
count_lean_multiple_backtrack = 0
whether_backtrack = False
tactic_list_at_top_per_key = self.tactic_list_tree[key][self.get_current_state_number(key)]
#print(f"key is {key}, we start to search")
if tactic_list_at_top_per_key == None:
assert False, f"tactic_list_at_top_per_key is None, key is {key}\ncurrent tactic list is {self.tactic_list_tree[key]}" \
f"\nwhether key in key_to_be_infered {key in self.key_to_be_infered}"
if len(tactic_list_at_top_per_key) != 0:
try:
#print(f'key is {key}, We apply tactic')
count_lean_tactic += 1
entered_tactic = self.revise_entered_tactic(tactic_list_at_top_per_key[0], key)
label_before_tactic = self.get_current_state_number(key)
#print(f'key is {key}, get proof before apply tactic is: {self.theorem_object_dict[key].get_current_lean_proof()}')
#print(f'key is {key}, current tactic_list_tree before apply tactic is {self.tactic_list_tree[key]}')
#print(f'key is {key}, entered_tactic is {entered_tactic}')
#print(f'key is {key}, current state before apply tactic is: {label_before_tactic}')
lean_output = self.theorem_object_dict[key].provide_tactic(entered_tactic[0], entered_tactic[1])
label_after_tactic = self.get_current_state_number(key)
#print(f'key is {key}, current state after apply tactic is: {label_after_tactic}')
if 'proof is complete' in lean_output[1]:
key_status = 'success'
print(f'key is {key}, search is success!')
print(f'key is {key}, the successful proof \n{self.theorem_object_dict[key].get_current_lean_proof()}')
#self.status_report()
self.parent_node_of_node[key][label_after_tactic] = label_before_tactic
self.prompts_tactic_state_list[key].append(
f"{entered_tactic[0]}\n{entered_tactic[1]}\n{lean_output[1]}")
self.tactic_list_tree[key][self.get_current_state_number(key)] = None
del self.tactic_list_tree[key][label_before_tactic][0]
#print(f'key is {key}, current tactic_list_tree after apply tactic is {self.tactic_list_tree[key]}')
self.check_path_length(key)
except Exception as e:
#print('tactic error happen')
#print(f'key is {key}, we apply tactic {entered_tactic} and see error')
#print(e)
#print(f'key is {key}, current state number after error after apply tactic is: {self.get_current_state_number(key)}')
#print(f'key is {key}, get proof after error after apply tactic is: {self.theorem_object_dict[key].get_current_lean_proof()}')
#print(f'key is {key}, current tactic_list_tree after error after apply tactic is {self.tactic_list_tree[key]}')
#print(f'key is {key}, we now delete the tactic from tactic list tree, the label is {self.get_current_state_number(key)}')
del self.tactic_list_tree[key][self.get_current_state_number(key)][0]
#print(f"key is {key}, tactic from tactic list tree is deleted, current tactic_list_tree is {self.tactic_list_tree[key]}")
self.check_path_length(key)
if self.check_if_failure_per_key(key):
key_status = 'failed'
print(f"key is {key}, tactic error then search failed!")
#self.status_report()
else:
if self.check_if_failure_per_key(key):
key_status = 'failed'
print(f"key is {key}, backtrack to zero and no tactic to try, search failed!")
#self.status_report()
print(f'key is {key}, backtrack phase activated')
whether_backtrack = True
count_lean_single_backtrack += 1
while True:
tactic_list_at_intermediate_node = self.tactic_list_tree[key][self.get_current_state_number(key)]
# print(f'tactic_tree_list: {self.tactic_list_tree[key]}')
# print(f"current state number: {self.get_current_state_number(key)}")
if len(tactic_list_at_intermediate_node) != 0:
break
if self.check_if_failure_per_key(key):
key_status = 'failed'
print(f"key is {key}, backtrack to zero and no tactic to try, search failed!")
#self.status_report()
break
#print(f'key is {key}, before back track step check length')
self.check_path_length(key)
#print(f'key is {key}, current state before backtrack is {self.get_current_state_number(key)}')
count_lean_multiple_backtrack += 1
lean_output = self.theorem_object_dict[key].do_back_track(self.back_track_tactic(key))
#print(f'current state after backtrack is {self.get_current_state_number(key)}')
#print('before delete the last ele of prompts_tactic_state_list')
#print('prompts_tactic_state_list is:')
#print(self.prompts_tactic_state_list[key])
#print(f'key for remove prompt here is {key}')
del self.prompts_tactic_state_list[key][-1]
#print('after back track step check length')
self.check_path_length(key)
except Exception as e:
print(f"key is {key}, exception happend")
print(e)
key_status = 'failed with error'
count_lean_tactic = 0
count_lean_single_backtrack = 0
count_lean_multiple_backtrack = 0
finally:
report_key_file = {}
report_key_file['count_lean_multiple_backtrack'] = count_lean_multiple_backtrack + count_lean_tactic
report_key_file['count_lean_single_backtrack'] = count_lean_single_backtrack + count_lean_tactic
report_key_file['count_lean_tactic_success'] = count_lean_tactic
report_key_file['key_status'] = key_status
report_key_file['tactic_list_tree'] = self.tactic_list_tree[key]
report_key_file['prompts_tactic_state_list'] = self.prompts_tactic_state_list[key]
report_key_file['theorem_object_dict'] = self.theorem_object_dict[key]
report_key_file['node_relation'] = self.parent_node_of_node[key]
report_key_file['whether_backtrack'] = whether_backtrack
pickle.dump(report_key_file, open(f'~/leandojo_project/atp_research/DFS/temp/temp_file_{key}_{self.round}_{self.experiment_id}.pkl','wb'))
def save_outcome(self):
counter_in_process = 0
counter_success = 0
counter_failed = 0
counter_too_long = 0
count_lean_single_backtrack = 0
count_lean_multiple_backtrack = 0
count_lean_tactic_success = 0
counter_failed_with_error = 0
proof_dict = {}
for key in tqdm(self.test_theorem_list,total=len(self.test_theorem_list),desc='saving results'):
count_lean_single_backtrack += self.count_lean_dict[key]['count_lean_single_backtrack']
count_lean_multiple_backtrack += self.count_lean_dict[key]['count_lean_multiple_backtrack']
count_lean_tactic_success += self.count_lean_dict[key]['count_lean_tactic_success']
proof_dict[key] = self.theorem_object_dict[key].get_current_lean_proof()
if self.root[key] == 'open':
counter_in_process += 1
if self.root[key] == 'success':
counter_success += 1
if self.root[key] == 'failed':
counter_failed += 1
if self.root[key] == 'failed, too long':
counter_too_long += 1
if self.root[key] == 'failed with error':
counter_failed_with_error += 1
self.counter_success = counter_success
self.counter_failed = counter_failed
self.counter_failed_with_error = counter_failed_with_error
self.counter_too_long = counter_too_long
if counter_success + counter_failed + counter_too_long + counter_failed_with_error + counter_in_process!= len(test_theorem_list):
assert False, 'number of theorm not equal to success, failed or too long, or in process'
outcome = {}
outcome['stats'] = {}
outcome['stats']['total_lean_count_single_backtrack'] = count_lean_single_backtrack
outcome['stats']['total_lean_count_multiple_backtrack'] = count_lean_multiple_backtrack
outcome['stats']['count_lean_tactic_success'] = count_lean_tactic_success
outcome['stats']['num_success'] = self.counter_success
outcome['stats']['num_failed'] = self.counter_failed
outcome['stats']['num_failed_with_error'] = self.counter_failed_with_error
outcome['stats']['num_too_long'] = self.counter_too_long
outcome['stats']['num_sampled_tactics'] = self.num_sampled_tactics
outcome['stats']['temperature'] = self.temperature
outcome['key_final_state'] = self.root
outcome['key_lean_count'] = self.count_lean_dict
outcome['key_proof'] = proof_dict
outcome['tactic_list_tree'] = self.tactic_list_tree
outcome['round_count'] = self.round_count
pickle.dump(outcome, open(self.saved_file_path, 'wb'))
if __name__ == '__main__':
'''os.environ["TOKENIZERS_PARALLELISM"] = "true"
if os.environ.get("TOKENIZERS_PARALLELISM") == "true":
print("TOKENIZERS_PARALLELISM is set to true")
else:
print("TOKENIZERS_PARALLELISM is not set to true")'''
parser = argparse.ArgumentParser(description='Description of your program.')
parser.add_argument('checkpoint_path', type=str, help='checkpoint_path')
parser.add_argument('number_of_gpu', type=int, help='number_of_gpu')
parser.add_argument('test_data_path', type=str, help='test_data_path')
parser.add_argument('saved_file_path', type=str, help='test_data_path')
parser.add_argument('max_workers', type=str, help='test_data_path')
parser.add_argument('num_sampled_tactics', type=str, help='test_data_path')
parser.add_argument('temperature', type=str, help='test_data_path')
parser.add_argument('num_test_theorem', type=str, help='test_data_path')
args = parser.parse_args()
checkpoint = args.checkpoint_path
number_of_gpu = args.number_of_gpu
test_data_path = args.test_data_path
saved_file_path = args.saved_file_path
max_workers = args.max_workers
num_sampled_tactics = args.num_sampled_tactics
temperature = args.temperature
num_test_theorem = args.num_test_theorem
swap_space = 100
print(f'checkpoint is {checkpoint}')
print(f'number_of_gpu is {number_of_gpu}')
print(f'test_data_path is {test_data_path}')
print(f'saved_file_path is {saved_file_path}')
print(f'num_test_theorem is {num_test_theorem}')
print(f'max_workers is {max_workers}')
print(f'number_sampled_tactic is {num_sampled_tactics}')
print(f'temperature is {temperature}')
print(f'swap_space is {swap_space}')
random.seed(42)
with open(test_data_path, 'r') as f:
test_theorem_list = json.load(f)
test_theorem_list = test_theorem_list[:int(num_test_theorem)]
#print(test_theorem_list)
#test_theorem_list = [9573315344600956080853155758,]
experiment_id = uuid.uuid4() # Generates a random UUID.
print(experiment_id)
llm = LLM(model=checkpoint, tensor_parallel_size=number_of_gpu, swap_space=swap_space)
evaluate_obj = DFS(num_sampled_tactics=int(num_sampled_tactics), temperature=float(temperature), test_theorem_list=test_theorem_list, max_workers=int(max_workers), saved_file_path=saved_file_path, experiment_id=experiment_id)
evaluate_obj.search()
print('Now we start saving')
evaluate_obj.save_outcome()
print('Now we finish saving. exit')
#command = "rm DFS/temp/temp_file*"
#result = subprocess.run(command, shell=True, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
|