-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathleanagent.py
1395 lines (1237 loc) · 59 KB
/
leanagent.py
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
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# import all the necessary libraries
import math
import ray
from collections import defaultdict
import os
import requests
import subprocess
import re
import shutil
from lean_dojo import *
import os
import json
import pickle
import numpy as np
from tqdm import tqdm
from loguru import logger
from lean_dojo import Theorem
from typing import List, Tuple, Optional
from lean_dojo import LeanGitRepo, Pos, is_available_in_cache
from lean_dojo import Theorem as LeanDojoTheorem
import json
import shutil
import random
from copy import copy
from pathlib import Path
from loguru import logger
from datetime import datetime, timedelta
from collections import defaultdict
from typing import Dict, List, Union
import generate_benchmark_lean4
import traceback
import sys
from tqdm import tqdm
from dynamic_database import *
import time
from pytorch_lightning.strategies import DDPStrategy
from prover.proof_search import Status, DistributedProver, SearchResult
import re
import lean_dojo
import pytorch_lightning as pl
from retrieval.model import PremiseRetriever
from retrieval.datamodule import RetrievalDataModule
from retrieval.main import run_cli
import torch
from pytorch_lightning.callbacks import ModelCheckpoint, EarlyStopping, LearningRateMonitor, Callback
from pytorch_lightning import seed_everything
# Set the seed for reproducibility
random.seed(3407) # https://arxiv.org/abs/2109.08203
BATCH_SIZE = 4
RAID_DIR = os.environ.get('RAID_DIR')
os.environ['RAY_TMPDIR'] = f"{RAID_DIR}/tmp"
repo_dir = f"{RAID_DIR}/repos_new"
DATA_DIR = "<DATA_DIR>"
CHECKPOINT_DIR = "<CHECKPOINT_DIR>"
EVAL_RESULTS_FILE_PATH = f"{RAID_DIR}/LeanAgent/<EVAL_RESULTS_FILE_PATH>"
DB_FILE_NAME = "<DB_FILE_NAME>"
PROOF_LOG_FILE_NAME = "proof_logs/<PROOF_LOG_FILE_NAME>"
ENCOUNTERED_THEOREMS_FILE = "<ENCOUNTERED_THEOREMS_FILE>"
FISHER_DIR = "<FISHER_DIR>" # Optional
repos_for_merged_dataset = []
repos_for_proving = []
# List of known repositories to process or skip
# Feel free to remove any repos from this list if you would like to test on them
known_repositories = [
"leanprover-community/mathlib4", # ReProver is trained on this
"leanprover-community/batteries", # functional programming instead of math
"leanprover-community/aesop",
"leanprover/lean4",
"leanprover-community/mathlib", # Mathlib3 version
"leanprover-community/mathlib3",
"leanprover/std4", # moved to batteries
"leanprover-community/duper", # functional programming instead of math
"leanprover/lake",
"openai/lean-gym",
"leanprover-community/lean4-metaprogramming-book",
"kmill/lean4-raytracer", # no theorems
"argumentcomputer/yatima", # trace problems
"ImperialCollegeLondon/formalising-mathematics-2024", # trace problems
"leanprover-community/ProofWidgets4", # trace problems
"leanprover/verso", # trace problems
"leanprover-community/NNG4", # trace problems
"ufmg-smite/lean-smt", # fails to trace due to windows-style line endings
"teorth/symmetric_project", # no compatible commit
"cmu-l3/llmlean", # irrelevant + only 4 theorems
"PatrickMassot/GlimpseOfLean", # strange trace problems with _parse_deps
"avigad/lamr", # trace problems
"leanprover-community/quote4", # no theorems
"leanprover-community/iris-lean", # trace problems
"aripiprazole/rinha", # incompatible commit
"leanprover/lean4-cli", # no theorems
"leanprover/LeanInk", # no theorems
"leanprover-community/lean-auto",
"leanprover-community/repl", # no theorems
"leanprover/doc-gen4", # no theorems
"leanprover/SampCert", # trace problems
"nomeata/loogle",
"risc0/risc0-lean4",
"PatrickMassot/verbose-lean4", # no theorems
"tydeu/lean4-alloy", # no theorems
"leanprover/leansat", # deprecated
"BoltonBailey/formal-snarks-project", # two theorems
"dwrensha/lean4-maze", # two theorems
"leanprover-community/mathport", # irrelevant
"argumentcomputer/LSpec", # one theorem
"reaslab/jixia", # no theorems
"riccardobrasca/flt3", # no theorems
"dwrensha/animate-lean-proofs", # irrelevant
"lean-ja/lean-by-example", # irrelevant
"NethermindEth/Clear", # no theorems
"fgdorais/lean4-parser", # irrelevant
"semorrison/lean-training-data", # irrelevant
"verse-lab/lean-ssr", # irrelevant
"GaloisInc/lean-llvm", # irrelevant
"argumentcomputer/Wasm.lean", # irrelevant
"NethermindEth/EVMYulLean", # irrelevant
"rwbarton/advent-of-lean-4", # irrelevant
"leanprover-community/tutorials4", # irrelevant
"haruhisa-enomoto/mathlib4-all-tactics", # irrelevant
"leanprover/LNSym",
"leanprover-community/flt-regular",
"opencompl/lean-mlir-old",
"rami3l/plfl",
"HEPLean/HepLean",
"forked-from-1kasper/ground_zero",
"verified-optimization/CvxLean",
"leanprover-community/sphere-eversion",
"optsuite/optlib",
"YaelDillies/LeanCamCombi",
"JamesGallicchio/LeanColls",
"T-Brick/c0deine",
"jjdishere/EG",
"alexkeizer/QpfTypes",
"fpvandoorn/LeanCourse23",
"marcusrossel/lean-egg",
"reilabs/proven-zk",
"algebraic-dev/soda",
"leanprover-community/llm",
"dignissimus/Untangle",
"argumentcomputer/Megaparsec.lean",
"emilyriehl/infinity-cosmos",
"BartoszPiotrowski/lean-premise-selection",
"djvelleman/HTPILeanPackage",
"girving/ray",
"Anderssorby/SDL.lean",
"pandaman64/lean-regex",
"brown-cs22/CS22-Lean-2023",
"hhu-adam/GameSkeleton",
"FR-vdash-bot/Algorithm",
"PeterKementzey/graph-library-for-lean4",
"arthurpaulino/LeanMySQL",
"arthurpaulino/NumLean",
"FormalSAT/trestle",
"nomeata/lean-wf-induct",
"leanprover/lean4checker",
"IPDSnelting/tba-2022",
"digama0/mm-lean4",
"KislyjKisel/Raylib.lean",
"algebraic-dev/melp",
"hhu-adam/Robo", # same as other tutorials but has lots of sorries
"hargoniX/socket.lean",
"kovach/etch",
"damek/gd-lean",
"0art0/lean-slides",
"forked-from-1kasper/lean4-categories",
"katydid/proofs",
"alexjbest/leaff",
"sinhp/Poly",
"lftcm2023/lftcm2023", # same as other tutorials but has lots of sorries
"lean-ja/lean99",
"leanprover/SHerLOC",
"Seasawher/mdgen",
"opencompl/egg-tactic-code",
"david-christiansen/ssft24",
"T-Brick/lean2wasm",
"hargoniX/cpdt-lean",
"jsm28/AperiodicMonotilesLean",
"draperlaboratory/ELFSage",
"rookie-joe/automatic-lean4-compilation",
"madvorak/fecssk",
"david-christiansen/bob24",
"awodey/joyal",
"BrownCS1951x/fpv2023", # same as other tutorials but has lots of sorries
"paulch42/lean-spec",
"siddhartha-gadgil/MetaExamples",
"dannypsnl/violet",
"arthurpaulino/LeanREPL",
"Kha/do-supplement",
"joehendrix/lean-sat-checker",
"ammkrn/timelib",
"kmill/LeanTeX",
"leanprover/lean4export",
"leanprover-community/mathlib3port",
"brown-cs22/CS22-Lean-2024", # same as other tutorials but has lots of sorries
"T-Brick/lean-wasm",
"crabbo-rave/Soup",
"argumentcomputer/RustFFI.lean",
"suhr/tmath",
"leanprover/leanbv",
"arthurpaulino/FxyLang",
"SchrodingerZhu/LeanGccBackend",
"lecopivo/lean4-karray",
"ImperialCollegeLondon/M1F-explained",
"proost-assistant/ProostLean",
"DavePearce/LeanEVM",
"algebraic-dev/ash",
"FormalizedFormalLogic/Arithmetization",
"cmu-l3/ntp-toolkit",
"dwrensha/tryAtEachStep",
"yangky11/lean4-example",
"T-Brick/DateTime",
"model-checking/rust-lean-models",
"MichaelStollBayreuth/EulerProducts",
"hargoniX/Flame",
"argumentcomputer/Http.lean",
"madvorak/vcsp",
"teorth/newton",
"apnelson1/Matroid",
"smorel394/TS1",
"ianjauslin-rutgers/pythagoras4",
"mortarsanjaya/IMOSLLean4",
"dupuisf/BibtexQuery",
"nomeata/lean-calcify",
"argumentcomputer/FFaCiL.lean",
"javra/iit",
"arthurpaulino/viper",
"lindy-labs/aegis",
"PatrickMassot/NNG4",
"argumentcomputer/YatimaStdLib.lean",
"fgdorais/lean4-unicode-basic",
"mhuisi/Uniq",
"Kha/macro-supplement",
"chenjulang/rubikcubegroup",
"arthurpaulino/LeanMusic",
"argumentcomputer/Ipld.lean",
"Odomontois/advent2022-lean",
"kbuzzard/IISc-experiments", # same as other tutorials but has lots of sorries
"ykonstant1/InfinitePrimes",
"alexkassil/natural_number_game_lean4",
"seewoo5/lean-poly-abc",
"rah4927/lean-dojo-mew",
"siddhartha-gadgil/proofs-and-programs-2023",
"PatrickMassot/lean4-game-server",
"knowsys/Formale-Systeme-in-LEAN", # same as other tutorials but has lots of sorries
"katydid/symbolic-automatic-derivatives",
"girving/interval",
"ImperialCollegeLondon/group-theory-experiments",
"knowsys/CertifyingDatalog",
"bergmannjg/leanCurl",
"vasnesterov/HadwigerNelson",
"FWuermse/lean-postgres",
"leanprover-community/import-graph",
"Human-Oriented-ATP/lean-tactics", # more about tactics than premises
"paulcadman/lean4-leetcode",
"argumentcomputer/Lurk.lean",
"AlexDuchnowski/rubiks-cube",
"SchrodingerZhu/lean-gccjit",
"JamesGallicchio/http",
"jtristan/UnicodeSkipListTableExample",
"adomani/MA4N1_2023", # same as other tutorials but has lots of sorries
"remimimimimi/leansec",
"hhu-adam/lean-i18n",
"RemyDegenne/testing-lower-bounds",
"mariainesdff/LocalClassFieldTheory",
"AviCraimer/relational-calculus-library-lean4",
"JLimperg/regensburg-itp-school-2023",
"jaalonso/Calculemus2",
"mseri/BET",
"xubaiw/Reservoir.lean",
"hargoniX/nest-core",
"siddhartha-gadgil/Polylean",
"MichaelStollBayreuth/Weights",
"sanchace/FRACTRAN",
"argumentcomputer/Poseidon.lean",
"madvorak/chomsky",
"T-Brick/ControlFlow",
"pa-ba/guarded-lean",
]
repos = []
lean_git_repos = []
personal_access_token = os.environ.get("GITHUB_ACCESS_TOKEN")
PR_TITLE = "[LeanAgent] Proofs"
PR_BODY = """
[LeanAgent](https://arxiv.org/abs/2410.06209) discovers a proof for a theorem with the `sorry` keyword.
---
<i>~LeanAgent - From the [LeanDojo](https://leandojo.org/) family</i>
"""
TMP_BRANCH = "_LeanAgent"
COMMIT_MESSAGE = "[LeanAgent] Proofs"
def clone_repo(repo_url):
"""Clone a git repository and return the path to the repository and its sha."""
repo_name = "/".join(repo_url.split('/')[-2:]).replace('.git', '')
logger.info(f"Cloning {repo_url}")
logger.info(f"Repo name: {repo_name}")
repo_name = repo_dir + "/" + repo_name
if os.path.exists(repo_name):
print(f"Deleting existing repository directory: {repo_name}")
shutil.rmtree(repo_name)
subprocess.run(["git", "clone", repo_url, repo_name])
process = subprocess.Popen(["git", "ls-remote", repo_url], stdout=subprocess.PIPE)
stdout, stderr = process.communicate()
sha = re.split(r'\t+', stdout.decode('utf-8'))[0]
return repo_name, sha
def branch_exists(repo_name, branch_name):
"""Check if a branch exists in a git repository."""
proc = subprocess.run(["git", "-C", repo_name, "branch", "-a"], capture_output=True, text=True)
branches = proc.stdout.split('\n')
local_branch = branch_name
remote_branch = f'remote/{branch_name}'
return any(branch.strip().endswith(local_branch) or branch.strip().endswith(remote_branch) for branch in branches)
def create_or_switch_branch(repo_name, branch_name, base_branch):
"""Create a branch in a git repository if it doesn't exist, or switch to it if it does."""
if not branch_exists(repo_name, branch_name):
subprocess.run(["git", "-C", repo_name, "checkout", "-b", branch_name], check=True)
else:
subprocess.run(["git", "-C", repo_name, "checkout", branch_name], check=True)
subprocess.run(["git", "-C", repo_name, "merge", base_branch, "-m", f"Merging {branch_name} into {base_branch}"], check=True)
def commit_changes(repo_name, commit_message):
"""Commit changes to a git repository."""
status = subprocess.run(["git", "-C", repo_name, "status", "--porcelain"], capture_output=True, text=True).stdout.strip()
if status == "":
print("No changes to commit.")
return False
subprocess.run(["git", "-C", repo_name, "add", "."], check=True)
subprocess.run(["git", "-C", repo_name, "commit", "-m", commit_message], check=True)
return True
def push_changes(repo_name, branch_name):
"""Push changes to a git repository."""
subprocess.run(["git", "-C", repo_name, "push", "-u", "origin", branch_name], check=True)
def get_default_branch(repo_full_name):
"""Get the default branch of a repository (default `main`)."""
url = f"https://api.github.com/repos/{repo_full_name}"
headers = {
"Authorization": f"token {personal_access_token}",
"Accept": "application/vnd.github.v3+json"
}
response = requests.get(url, headers=headers)
if response.status_code == 200:
return response.json()['default_branch']
else:
logger.info(f"Failed to get default branch for {repo_full_name}")
return "main"
def create_pull_request(repo_full_name, title, body, head_branch):
"""Create a pull request in a repository."""
base_branch = get_default_branch(repo_full_name)
url = f"https://api.github.com/repos/{repo_full_name}/pulls"
headers = {
"Authorization": f"token {personal_access_token}",
"Accept": "application/vnd.github.v3+json"
}
data = {
"title": title,
"body": body,
"head": head_branch,
"base": base_branch
}
response = requests.post(url, headers=headers, json=data)
if response.status_code == 201:
print("Pull request created successfully: " + response.json()['html_url'])
return response.json()['html_url']
else:
print("Failed to create pull request", response.text)
return ""
def get_compatible_commit(url):
"""Find the most recent commit with a Lean version that LeanAgent supports."""
try:
process = subprocess.Popen(["git", "ls-remote", url], stdout=subprocess.PIPE)
stdout, stderr = process.communicate()
latest_commit = re.split(r'\t+', stdout.decode('utf-8'))[0]
logger.info(f"Latest commit: {latest_commit}")
new_url = url.replace('.git', '')
logger.info(f"Creating LeanGitRepo for {new_url}")
repo = LeanGitRepo(new_url, latest_commit)
logger.info(f"Getting config for {url}")
config = repo.get_config("lean-toolchain")
v = generate_benchmark_lean4.get_lean4_version_from_config(config["content"])
if generate_benchmark_lean4.is_supported_version(v):
logger.info(f"Latest commit compatible for url {url}")
return latest_commit, v
logger.info(f"Searching for compatible commit for {url}")
try:
subprocess.run(["git", "rev-parse", "--is-inside-work-tree"],
check=True,
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL)
logger.info("Already in a Git repository")
except subprocess.CalledProcessError:
logger.info("Not in a Git repository. Initializing one.")
subprocess.run(["git", "init"], check=True)
process = subprocess.Popen(
["git", "fetch", "--depth=1000000", url], # Fetch commits
stdout=subprocess.PIPE,
stderr=subprocess.PIPE
)
logger.info(f"Fetching commits for {url}")
_, stderr = process.communicate()
if process.returncode != 0:
raise Exception(f"Git fetch command failed: {stderr.decode('utf-8')}")
logger.info(f"Fetched commits for {url}")
process = subprocess.Popen(
["git", "log", "--format=%H", "FETCH_HEAD"], # Get list of commits
stdout=subprocess.PIPE,
stderr=subprocess.PIPE
)
logger.info(f"Getting list of commits for {url}")
stdout, stderr = process.communicate()
if process.returncode != 0:
raise Exception(f"Git log command failed: {stderr.decode('utf-8')}")
commits = stdout.decode('utf-8').strip().split('\n')
logger.info(f"Found {len(commits)} commits for {url}")
for commit in commits:
new_url = url.replace('.git', '')
repo = LeanGitRepo(new_url, commit)
config = repo.get_config("lean-toolchain")
v = generate_benchmark_lean4.get_lean4_version_from_config(config["content"])
if generate_benchmark_lean4.is_supported_version(v):
logger.info(f"Found compatible commit {commit} for {url}")
return commit, v
raise Exception("No compatible commit found")
except Exception as e:
logger.info(f"Error in get_compatible_commit: {str(e)}")
return None, None
def find_and_save_compatible_commits(repo_info_file, lean_git_repos):
"""Finds compatible commits for various repositories"""
updated_repos = []
for repo in lean_git_repos:
url = repo.url
if not url.endswith('.git'):
url = url + '.git'
sha = None
v = None
if "mathlib4" in url:
sha = "2b29e73438e240a427bcecc7c0fe19306beb1310"
v = "v4.8.0"
elif "SciLean" in url:
sha = "22d53b2f4e3db2a172e71da6eb9c916e62655744"
v = "v4.7.0"
elif "pfr" in url:
sha = "fa398a5b853c7e94e3294c45e50c6aee013a2687"
v = "v4.8.0-rc1"
else:
sha, v = get_compatible_commit(url)
if not sha:
logger.info(f"Failed to find a compatible commit for {url}")
continue
updated_repos.append({"url": url.replace('.git', ''), "commit": sha, "version": v})
with open(repo_info_file, 'w') as f:
json.dump(updated_repos, f)
return updated_repos
def search_github_repositories(language="Lean", num_repos=10):
"""Search for the given number of repositories on GitHub that have the given language."""
headers = {'Authorization': personal_access_token}
query_params = {
'q': f'language:{language}',
'sort': 'stars',
'order': 'desc',
'per_page': 100,
}
cloned_count = 0
page = 1
while cloned_count < num_repos:
query_params['page'] = page
response = requests.get('https://api.github.com/search/repositories', headers=headers, params=query_params)
if response.status_code == 200:
repositories = response.json()['items']
for repo in repositories:
if cloned_count >= num_repos:
break
repo_full_name = repo['full_name']
logger.info(f"Processing {repo_full_name}")
if repo_full_name not in known_repositories:
name = None
try:
clone_url = repo['clone_url']
repo_name, sha = clone_repo(clone_url)
name = repo_name
url = clone_url.replace('.git', '')
lean_git_repo = LeanGitRepo(url, sha)
lean_git_repos.append(lean_git_repo)
repos.append(repo_full_name)
cloned_count += 1
logger.info(f"Cloned {repo_full_name}")
except Exception as e:
shutil.rmtree(name)
logger.info(f"Failed to clone {repo_full_name} because of {e}")
else:
logger.info(f"Skipping {repo_full_name} since it is a known repository")
page += 1
else:
logger.info("Failed to search GitHub", response.status_code)
break
# Check if we've reached the end of the search results
if len(repositories) < 100:
break
logger.info(f"Total repositories processed: {cloned_count}")
def _eval(data, preds_map) -> Tuple[float, float, float]:
"""Evaluates the retrieval model."""
R1 = []
R10 = []
MRR = []
for thm in tqdm(data):
for i, _ in enumerate(thm["traced_tactics"]):
pred = None
key = (thm["file_path"], thm["full_name"], tuple(thm["start"]), i)
if key in preds_map:
pred = preds_map[key]
else:
continue
all_pos_premises = set(pred["all_pos_premises"])
if len(all_pos_premises) == 0:
continue
retrieved_premises = pred["retrieved_premises"]
TP1 = retrieved_premises[0] in all_pos_premises
R1.append(float(TP1) / len(all_pos_premises))
TP10 = len(all_pos_premises.intersection(retrieved_premises[:10]))
R10.append(float(TP10) / len(all_pos_premises))
for j, p in enumerate(retrieved_premises):
if p in all_pos_premises:
MRR.append(1.0 / (j + 1))
break
else:
MRR.append(0.0)
R1 = 100 * np.mean(R1)
R10 = 100 * np.mean(R10)
MRR = np.mean(MRR)
return R1, R10, MRR
def load_fisher_information(file_path):
"""Loads the Fisher Information Matrix."""
try:
with open(file_path, 'rb') as f:
fisher_info = pickle.load(f)
logger.info("Fisher Information successfully loaded.")
return fisher_info
except FileNotFoundError:
logger.error(f"No Fisher Information file found at {file_path}.")
return None
def find_latest_checkpoint():
"""Finds the most recent checkpoint."""
checkpoint_dir = RAID_DIR + "/" + CHECKPOINT_DIR
all_checkpoints = [os.path.join(checkpoint_dir, f) for f in os.listdir(checkpoint_dir) if f.endswith(".ckpt")]
if not all_checkpoints:
raise FileNotFoundError("No checkpoints found.")
latest_checkpoint = max(all_checkpoints, key=os.path.getmtime)
logger.info(f"Using the latest checkpoint: {latest_checkpoint}")
return latest_checkpoint
def find_latest_fisher():
"""Finds the most recent Fisher Information Matrix."""
fisher_dir = RAID_DIR + "/" + FISHER_DIR
all_fisher = [os.path.join(fisher_dir, f) for f in os.listdir(fisher_dir) if f.endswith(".pkl")]
if not all_fisher:
raise FileNotFoundError("No Fisher Information Matrices found.")
latest_fisher = max(all_fisher, key=os.path.getmtime)
logger.info(f"Using the latest Fisher Information Matrix: {latest_fisher}")
return latest_fisher
def theorem_identifier(theorem: Theorem) -> Tuple[str, str, Tuple[int, int], Tuple[int, int]]:
"""Returns a unique identifier for a theorem."""
return (theorem.full_name, str(theorem.file_path), tuple(theorem.start), tuple(theorem.end))
def process_theorem_batch(theorem_batch, positions_batch, repo, db, prover, dynamic_database_json_path):
"""Processes a batch of theorems."""
lean_dojo_theorems = [t[1] for t in theorem_batch]
results = prover.search_unordered(LeanGitRepo(repo.url, repo.commit), lean_dojo_theorems, positions_batch)
# Create a mapping from LeanDojoTheorem to our Theorem
theorem_map = {ldj_thm: thm for thm, ldj_thm in theorem_batch}
for result in results:
if isinstance(result, SearchResult):
if result.theorem in theorem_map:
theorem = theorem_map[result.theorem]
if result.status == Status.PROVED:
logger.info(f"Proof found for {theorem.full_name}")
traced_tactics = [
AnnotatedTactic(
tactic=tactic,
annotated_tactic=(tactic, []),
state_before="",
state_after=""
) for tactic in result.proof
]
theorem.traced_tactics = traced_tactics
repo.change_sorry_to_proven(theorem, PROOF_LOG_FILE_NAME)
db.update_repository(repo)
logger.info(f"Updated theorem {theorem.full_name} in the database")
else:
logger.info(f"No proof found for {theorem.full_name}")
else:
logger.warning(f"Theorem not found in theorem_map: {result.theorem}")
else:
logger.warning(f"Unexpected result type")
db.to_json(dynamic_database_json_path)
def save_progress(all_encountered_theorems):
"""Saves the set of encountered theorems."""
logger.info("Saving encountered theorems...")
with open(ENCOUNTERED_THEOREMS_FILE, 'wb') as f:
pickle.dump(all_encountered_theorems, f)
def load_encountered_theorems(file_path):
"""Loads the theorems that have been encountered."""
all_encountered_theorems = set()
if os.path.exists(file_path):
try:
with open(file_path, 'rb') as f:
file_content = f.read()
if file_content: # Check if the file is not empty
all_encountered_theorems = pickle.loads(file_content)
else:
logger.warning(f"The file {file_path} is empty. Starting with an empty set.")
except (EOFError, pickle.UnpicklingError) as e:
logger.warning(f"Error reading {file_path}: {e}. Starting with an empty set.")
except Exception as e:
logger.error(f"Unexpected error when reading {file_path}: {e}. Starting with an empty set.")
else:
logger.info(f"The file {file_path} does not exist. Starting with an empty set.")
return all_encountered_theorems
def prove_sorry_theorems(db: DynamicDatabase, prover: DistributedProver, dynamic_database_json_path, repos_to_include: Optional[List[Tuple[str, str]]] = None, batch_size: int = 12):
"""Proves sorry theorems."""
repos_to_process = db.repositories if repos_to_include is None else [
repo for repo in db.repositories if (repo.url, repo.commit) in repos_to_include
]
# To avoid proving the same theorem multiple times, potentially from different versions of the
# same repo, we sort the repositories
repos_to_process.sort(key=lambda r: r.metadata['date_processed'], reverse=True)
processed_theorems: Set[Tuple[str, str, Tuple[int, int], Tuple[int, int]]] = set()
all_encountered_theorems: Set[Tuple[str, str, Tuple[int, int], Tuple[int, int]]] = set()
last_save_time = datetime.datetime.now()
save_interval = timedelta(minutes=30)
# Load previously encountered theorems
all_encountered_theorems = load_encountered_theorems(ENCOUNTERED_THEOREMS_FILE)
for repo in repos_to_process:
sorry_theorems = repo.sorry_theorems_unproved
repo_url = repo.url
repo_commit = repo.commit
logger.info(f"Found {len(sorry_theorems)} sorry theorems to prove")
theorem_batch = []
positions_batch = []
for theorem in tqdm(sorry_theorems, desc=f"Processing theorems from {repo.name}", unit="theorem"):
# Ignore sorry theorems from the repo's dependencies
if theorem.url != repo_url or theorem.commit != repo_commit:
continue
theorem_id = theorem_identifier(theorem)
if theorem_id in all_encountered_theorems:
logger.info(f"Skipping already encountered theorem: {theorem.full_name}")
continue
all_encountered_theorems.add(theorem_id)
if theorem_id in processed_theorems:
logger.info(f"Skipping already processed theorem: {theorem.full_name}")
continue
processed_theorems.add(theorem_id)
logger.info(f"Searching for proof for {theorem.full_name}")
logger.info(f"Position: {theorem.start}")
# Convert our Theorem to LeanDojo Theorem
lean_dojo_theorem = LeanDojoTheorem(
repo=LeanGitRepo(repo_url, repo_commit),
file_path=theorem.file_path,
full_name=theorem.full_name
)
theorem_batch.append((theorem, lean_dojo_theorem))
positions_batch.append(Pos(*theorem.start))
if len(theorem_batch) == batch_size:
process_theorem_batch(theorem_batch, positions_batch, repo, db, prover, dynamic_database_json_path)
theorem_batch = []
positions_batch = []
current_time = datetime.datetime.now()
if current_time - last_save_time >= save_interval:
save_progress(all_encountered_theorems)
last_save_time = current_time
# Process any remaining theorems in the last batch
if theorem_batch:
process_theorem_batch(theorem_batch, positions_batch, repo, db, prover, dynamic_database_json_path)
save_progress(all_encountered_theorems)
logger.info("Finished attempting to prove sorry theorems")
def add_repo_to_database(dynamic_database_json_path, repo, db):
"""Adds a repository to the dynamic database."""
# Prepare the data necessary to add this repo to the dynamic database
url = repo.url
if not url.endswith('.git'):
url = url + '.git'
logger.info(f"Processing {url}")
if "mathlib4" in url:
sha = "2b29e73438e240a427bcecc7c0fe19306beb1310"
v = "v4.8.0"
elif "SciLean" in url:
sha = "22d53b2f4e3db2a172e71da6eb9c916e62655744"
v = "v4.7.0"
elif "pfr" in url:
sha = "fa398a5b853c7e94e3294c45e50c6aee013a2687"
v = "v4.8.0-rc1"
else:
sha, v = get_compatible_commit(url)
if not sha:
logger.info(f"Failed to find a compatible commit for {url}")
return None
logger.info(f"Found compatible commit {sha} for {url}")
logger.info(f"Lean version: {v}")
url = url.replace('.git', '')
repo = LeanGitRepo(url, sha)
dir_name = repo.url.split("/")[-1] + "_" + sha
dst_dir = RAID_DIR + "/" + DATA_DIR + "/" + dir_name
logger.info(f"Generating benchmark at {dst_dir}")
traced_repo, _, _, total_theorems = generate_benchmark_lean4.main(repo.url, sha, dst_dir)
if not traced_repo:
logger.info(f"Failed to trace {url}")
return None
if total_theorems < 3 * BATCH_SIZE: # Should be enough theorems for train/val/test
logger.info(f"No theorems found in {url}")
return None
logger.info(f"Finished generating benchmark at {dst_dir}")
# Add the new repo to the dynamic database
config = repo.get_config("lean-toolchain")
v = generate_benchmark_lean4.get_lean4_version_from_config(config["content"])
theorems_folder = dst_dir + "/random"
premise_files_corpus = dst_dir + "/corpus.jsonl"
files_traced = dst_dir + "/traced_files.jsonl"
pr_url = None
data = {
"url": repo.url,
"name": "/".join(repo.url.split("/")[-2:]),
"commit": repo.commit,
"lean_version": v,
"lean_dojo_version": lean_dojo.__version__,
"metadata": {
"date_processed": datetime.datetime.now(),
},
"theorems_folder": theorems_folder,
"premise_files_corpus": premise_files_corpus,
"files_traced": files_traced,
"pr_url": pr_url
}
repo = Repository.from_dict(data)
logger.info("Before adding new repo:")
db.print_database_contents()
db.add_repository(repo)
logger.info("After adding new repo:")
db.print_database_contents()
db.to_json(dynamic_database_json_path)
return "Done"
def replace_sorry_with_proof(proofs):
"""Replace the `sorry` with the proof text in the Lean files."""
logger.info(f"Replacing sorries with {len(proofs)} proofs!")
# Group proofs by file paths
proofs_by_file = {}
for proof in proofs:
file_path, start, end, proof_text, theorem_name = proof
if file_path not in proofs_by_file:
proofs_by_file[file_path] = []
proofs_by_file[file_path].append((start, end, proof_text))
for file_path, proofs in proofs_by_file.items():
with open(file_path, 'r') as file:
lines = file.readlines()
# sort proof by starting line and column number (working bottom up retains positions)
proofs.sort(key=lambda x: (x[0].line_nb, x[0].column_nb), reverse=True)
for start, end, proof_text in proofs:
start_line, start_col = start.line_nb - 1, start.column_nb - 1
end_line, end_col = end.line_nb - 1, end.column_nb - 1
original_text = ''.join(lines[start_line:end_line + 1])
new_text = original_text.replace('sorry', proof_text, 1)
lines[start_line:end_line + 1] = new_text
with open(file_path, 'w') as file:
file.writelines(lines)
logger.info("Finished replacing sorries with proofs!")
def calculate_difficulty(theorem: Theorem) -> Union[float, None]:
"""Calculates the difficulty of a theorem."""
proof_steps = theorem.traced_tactics
if any('sorry' in step.tactic for step in proof_steps):
return float('inf') # Hard (no proof)
if len(proof_steps) == 0:
return None # To be distributed later
return math.exp(len(proof_steps))
def categorize_difficulty(difficulty: Union[float, None], percentiles: List[float]) -> str:
"""Categorizes the difficulty of a theorem."""
if difficulty is None:
return "To_Distribute"
if difficulty == float('inf'):
return "Hard (No proof)"
elif difficulty <= percentiles[0]:
return "Easy"
elif difficulty <= percentiles[1]:
return "Medium"
else:
return "Hard"
def sort_repositories_by_difficulty(db: DynamicDatabase) -> List[Repository]:
"""Sorts repositories by the difficulty of their theorems."""
difficulties_by_repo = defaultdict(list)
all_difficulties = []
print("Ready to calculate difficulties of all theorems")
for repo in db.repositories:
print(f"Starting {repo.name}")
for theorem in repo.get_all_theorems:
difficulty = calculate_difficulty(theorem)
theorem.difficulty_rating = difficulty
difficulties_by_repo[repo].append((theorem.full_name, str(theorem.file_path), tuple(theorem.start), tuple(theorem.end), difficulty))
if difficulty is not None:
all_difficulties.append(difficulty)
db.update_repository(repo)
print(f"Finished {repo.name}")
percentiles = np.percentile(all_difficulties, [33, 67])
categorized_theorems = defaultdict(lambda: defaultdict(list))
print("Ready to categorize theorems")
for repo, theorems in difficulties_by_repo.items():
print(f"Starting {repo.name}")
for theorem_name, file_path, start, end, difficulty in theorems:
category = categorize_difficulty(difficulty, percentiles)
categorized_theorems[repo][category].append((theorem_name, file_path, start, end, difficulty))
print(f"Finished {repo.name}")
print("Distributed theorems with no proofs")
for repo in categorized_theorems:
print(f"Starting {repo.name}")
to_distribute = categorized_theorems[repo]["To_Distribute"]
chunk_size = len(to_distribute) // 3
for i, category in enumerate(["Easy", "Medium", "Hard"]):
start = i * chunk_size
end = start + chunk_size if i < 2 else None
categorized_theorems[repo][category].extend(to_distribute[start:end])
del categorized_theorems[repo]["To_Distribute"]
print(f"Finished {repo.name}")
# Sort repositories based on the number of easy theorems
sorted_repos = sorted(categorized_theorems.keys(), key=lambda r: len(categorized_theorems[r]["Easy"]), reverse=True)
return sorted_repos, categorized_theorems, percentiles
def save_sorted_repos(sorted_repos: List[Repository], file_path: str):
"""Saves the sorted repositories to a file."""
sorted_repo_data = [
{
"url": repo.url,
"commit": repo.commit,
"name": repo.name
} for repo in sorted_repos
]
with open(file_path, 'w') as f:
json.dump(sorted_repo_data, f, indent=2)
def load_sorted_repos(file_path: str) -> List[Tuple[str, str, str]]:
"""Loads the sorted repositories from a file."""
with open(file_path, 'r') as f:
sorted_repo_data = json.load(f)
return [(repo["url"], repo["commit"], repo["name"]) for repo in sorted_repo_data]
def write_skip_file(repo_url):
"""Writes a repository URL to a file to skip it."""
skip_file_path = os.path.join(RAID_DIR, DATA_DIR, "skip_repo.txt")
with open(skip_file_path, 'w') as f:
f.write(repo_url)
def should_skip_repo():
"""Checks if a repository should be skipped."""
skip_file_path = os.path.join(RAID_DIR, DATA_DIR, "skip_repo.txt")
if os.path.exists(skip_file_path):
with open(skip_file_path, 'r') as f:
repo_url = f.read().strip()
return True, repo_url
return False, None
def main():
"""
Main function to run LeanAgent.
"""
global repos_for_merged_dataset
global repos_for_proving
global lean_git_repos
try:
current_epoch = 0
epochs_per_repo = 1
run_progressive_training = True
use_fisher = False
single_repo = True
curriculum_learning = True
num_repos = 15
dynamic_database_json_path = RAID_DIR + "/" + DB_FILE_NAME
lambdas = None
if run_progressive_training:
logger.info("Running progressive training")
lambdas = [0.1]
else:
logger.info("Running retrieval baseline")
lambdas = [0.0]
# Add debug information
logger.info("Configuring LeanDojo...")
generate_benchmark_lean4.configure_leandojo()
logger.info("LeanDojo configured")
# Check if the current process is the main one
is_main_process = int(os.environ.get('LOCAL_RANK', '0')) == 0
# Initialize the database if it doesn't exist or is empty
if is_main_process:
logger.info("Starting the main process")
if not os.path.exists(dynamic_database_json_path) or os.path.getsize(dynamic_database_json_path) == 0:
# File doesn't exist or is empty, initialize it
logger.info(f"Initializing new database at {dynamic_database_json_path}")
db = DynamicDatabase()
db.to_json(dynamic_database_json_path)
else:
try:
logger.info(f"Loading database from {dynamic_database_json_path}")
db = DynamicDatabase.from_json(dynamic_database_json_path)
logger.info(f"Loaded database from {dynamic_database_json_path}")
except json.JSONDecodeError:
# If there's an error decoding the JSON, initialize a new database
logger.warning(f"Error decoding JSON from {dynamic_database_json_path}. Initializing new database.")
db = DynamicDatabase()
db.to_json(dynamic_database_json_path)
logger.info(f"Found {num_repos} repositories")
# If curriculum learning is enabled, initialize repositories and sort them by difficulty
if curriculum_learning:
logger.info("Starting curriculum learning")