3 # Any copyright is dedicated to the Public Domain.
4 # https://creativecommons.org/publicdomain/zero/1.0/
6 # Written by Francois Fleuret <francois@fleuret.org>
8 import math, sys, argparse, time, tqdm, os, datetime, warnings
10 import torch, torchvision
12 from torch.nn import functional as F
14 # torch.autograd.set_detect_anomaly(True) #!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
17 import mygpt, tasks, problems
19 ######################################################################
24 if x in {"1", "true", "yes"}:
26 elif x in {"0", "false", "no"}:
32 parser = argparse.ArgumentParser(
33 description="An implementation of GPT with cache.",
34 formatter_class=argparse.ArgumentDefaultsHelpFormatter,
41 help="byheart, learnop, guessop, mixing, memory, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl, grid, qmlp",
44 parser.add_argument("--log_filename", type=str, default="train.log", help=" ")
46 parser.add_argument("--result_dir", type=str, default=None)
48 parser.add_argument("--seed", type=int, default=0)
50 parser.add_argument("--max_percents_of_test_in_train", type=int, default=1)
52 parser.add_argument("--force_cpu", type=str2bool, default=False)
54 ########################################
56 parser.add_argument("--nb_epochs", type=int, default=25)
58 parser.add_argument("--physical_batch_size", type=int, default=None)
60 parser.add_argument("--batch_size", type=int, default=25)
62 parser.add_argument("--nb_train_samples", type=int, default=None)
64 parser.add_argument("--nb_test_samples", type=int, default=None)
66 parser.add_argument("--optim", type=str, default="adam")
68 ########################################
70 parser.add_argument("--nb_warmup_iter", type=int, default=100)
72 parser.add_argument("--nb_decay_iter", type=int, default=5000)
74 parser.add_argument("--learning_rate", type=float, default=6e-4)
76 parser.add_argument("--min_learning_rate", type=float, default=6e-5)
80 parser.add_argument("--legacy_lr_schedule", type=str2bool, default=True)
82 parser.add_argument("--legacy_large_lr", type=float, default=1e-4)
84 parser.add_argument("--legacy_small_lr", type=float, default=2e-5)
86 parser.add_argument("--legacy_nb_epoch_large_lr", type=float, default=10)
88 ########################################
90 parser.add_argument("--model", type=str, default=None)
92 parser.add_argument("--attention", type=str, default=None)
94 parser.add_argument("--memex_proba", type=float, default=0)
96 parser.add_argument("--memex_nb_epochs", type=float, default=None)
98 parser.add_argument("--dim_model", type=int, default=None)
100 parser.add_argument("--dim_keys", type=int, default=None)
102 parser.add_argument("--dim_hidden", type=int, default=None)
104 parser.add_argument("--nb_heads", type=int, default=None)
106 parser.add_argument("--nb_lines", type=int, default=None)
108 parser.add_argument("--caterpillar_height", type=int, default=None)
110 parser.add_argument("--gate_dropout_proba", type=float, default=0.0)
112 parser.add_argument("--gate_dropout_sync", type=str2bool, default=False)
114 parser.add_argument("--gate_dropout_replace", type=str2bool, default=False)
116 parser.add_argument("--rho_inner_loss", type=float, default=0.0)
118 parser.add_argument("--nb_blocks", type=int, default=None)
120 parser.add_argument("--dropout", type=float, default=0.1)
122 ########################################
124 parser.add_argument("--deterministic_synthesis", action="store_true", default=False)
126 parser.add_argument("--no_checkpoint", action="store_true", default=False)
128 parser.add_argument("--continue_training", action="store_true", default=False)
130 parser.add_argument("--checkpoint_name", type=str, default="checkpoint.pth")
132 ##############################
135 parser.add_argument("--rpl_nb_starting_values", type=int, default=3)
137 parser.add_argument("--rpl_max_input", type=int, default=9)
139 parser.add_argument("--rpl_prog_len", type=int, default=8)
141 parser.add_argument("--rpl_nb_runs", type=int, default=5)
143 parser.add_argument("--rpl_no_prog", action="store_true", default=False)
145 ##############################
148 parser.add_argument("--grid_size", type=int, default=6)
150 parser.add_argument("--grid_nb_colors", type=int, default=6)
152 parser.add_argument("--grid_nb_shapes", type=int, default=6)
154 ##############################
157 parser.add_argument("--picoclvr_nb_colors", type=int, default=5)
159 parser.add_argument("--picoclvr_height", type=int, default=12)
161 parser.add_argument("--picoclvr_width", type=int, default=16)
163 parser.add_argument("--picocvlr_prune_properties", type=str, default="none")
165 ##############################
168 parser.add_argument("--maze_height", type=int, default=13)
170 parser.add_argument("--maze_width", type=int, default=21)
172 parser.add_argument("--maze_nb_walls", type=int, default=15)
174 ##############################
177 parser.add_argument("--snake_height", type=int, default=9)
179 parser.add_argument("--snake_width", type=int, default=12)
181 parser.add_argument("--snake_nb_colors", type=int, default=5)
183 parser.add_argument("--snake_length", type=int, default=200)
185 ##############################
188 parser.add_argument("--stack_nb_steps", type=int, default=100)
190 parser.add_argument("--stack_nb_stacks", type=int, default=3)
192 parser.add_argument("--stack_nb_digits", type=int, default=3)
194 parser.add_argument("--stack_fraction_values_for_train", type=float, default=0.75)
196 ##############################
199 parser.add_argument("--expr_nb_variables", type=int, default=5)
201 parser.add_argument("--expr_sequence_length", type=int, default=40)
203 parser.add_argument("--expr_operand_max", type=int, default=9)
205 parser.add_argument("--expr_result_max", type=int, default=99)
207 parser.add_argument("--expr_input_file", type=str, default=None)
209 ##############################
212 parser.add_argument("--memory_len_total", type=int, default=32)
214 ##############################
217 parser.add_argument("--mixing_hard", action="store_true", default=False)
219 parser.add_argument("--mixing_deterministic_start", action="store_true", default=False)
221 ######################################################################
223 # args = parser.parse_args()
225 args, sup_args = parser.parse_known_args()
227 sup_args = dict([x.removeprefix("--").split("=") for x in sup_args])
229 if args.result_dir is None:
230 args.result_dir = f"results_{args.task}_{args.model}"
232 ######################################################################
234 if not args.force_cpu and torch.cuda.is_available():
235 device = torch.device("cuda")
236 torch.backends.cuda.matmul.allow_tf32 = True
238 device = torch.device("cpu")
240 ######################################################################
242 default_task_args = {
245 "physical_batch_size": 25,
246 "nb_train_samples": 250000,
247 "nb_test_samples": 10000,
251 "physical_batch_size": 25,
252 "nb_train_samples": 50000,
253 "nb_test_samples": 10000,
257 "physical_batch_size": 25,
258 "nb_train_samples": 2500000,
259 "nb_test_samples": 10000,
263 "physical_batch_size": 25,
264 "nb_train_samples": 250000,
265 "nb_test_samples": 10000,
269 "physical_batch_size": 10,
270 "nb_train_samples": 100000,
271 "nb_test_samples": 1000,
275 "physical_batch_size": 25,
276 "nb_train_samples": 1000000,
277 "nb_test_samples": 10000,
281 "physical_batch_size": 25,
282 "nb_train_samples": 50000,
283 "nb_test_samples": 10000,
287 "physical_batch_size": 5,
288 "nb_train_samples": 100000,
289 "nb_test_samples": 10000,
293 "physical_batch_size": 25,
294 "nb_train_samples": 250000,
295 "nb_test_samples": 10000,
299 "physical_batch_size": 5,
300 "nb_train_samples": 2500000,
301 "nb_test_samples": 10000,
305 "physical_batch_size": 25,
306 "nb_train_samples": 250000,
307 "nb_test_samples": 10000,
311 "physical_batch_size": 25,
312 "nb_train_samples": 100000,
313 "nb_test_samples": 1000,
317 "physical_batch_size": 25,
318 "nb_train_samples": 50000,
319 "nb_test_samples": 10000,
323 "physical_batch_size": 25,
324 "nb_train_samples": 25000,
325 "nb_test_samples": 10000,
329 "physical_batch_size": 25,
330 "nb_train_samples": 250000,
331 "nb_test_samples": 10000,
335 "physical_batch_size": 5,
336 "nb_train_samples": 60000,
337 "nb_test_samples": 10000,
341 if args.task in default_task_args:
342 for k, v in default_task_args[args.task].items():
343 if getattr(args, k) is None:
346 ######################################################################
348 default_model_args = {
358 "attention": "caterpillar",
364 "caterpillar_height": 4,
376 "attention": "caterpillar",
382 "caterpillar_height": 4,
394 "attention": "caterpillar",
400 "caterpillar_height": 32,
412 "attention": "caterpillar",
429 "attention": "caterpillar",
439 if args.model in default_model_args:
440 for k, v in default_model_args[args.model].items():
441 if getattr(args, k) is None:
444 raise ValueError(f"Unknown model {args.model}")
446 ######################################################################
449 os.mkdir(args.result_dir)
450 except FileExistsError:
451 if not args.continue_training:
452 print(f"result directory {args.result_dir} already exists")
455 loss_file = open(os.path.join(args.result_dir, "loss.dat"), "a")
457 log_file = open(os.path.join(args.result_dir, args.log_filename), "a")
460 # torch.backends.cudnn.deterministic = True
461 # torch.backends.cudnn.benchmark = False
462 # torch.use_deterministic_algorithms(True)
463 torch.manual_seed(args.seed)
464 if torch.cuda.is_available():
465 torch.cuda.manual_seed_all(args.seed)
467 ######################################################################
471 t = time.strftime("%Y%m%d-%H:%M:%S ", time.localtime())
473 if log_file is not None:
474 log_file.write(t + s + "\n")
481 with os.popen("sha256sum *.py") as f:
483 log_string(f"sha256sum {l.strip()}")
485 now = time.strftime("%Y%m%d-%H%M%S", time.localtime())
486 os.system(f"tar zcvf {args.result_dir}/src-{now}.tgz *.py *.sh")
488 log_string(f"argv {' '.join(sys.argv)}")
491 log_string(f"args.{n} {getattr(args, n)}")
493 for k, v in sup_args.items():
494 log_string(f'sup_args["{k}"] "{v}"')
497 ######################################################################
500 def get_lr(n_epoch, it):
501 if args.legacy_lr_schedule:
502 # my crude scheduling to compare to previous baseline, added
505 if it < args.nb_warmup_iter:
506 return args.legacy_large_lr * it / args.nb_warmup_iter
507 elif n_epoch < args.legacy_nb_epoch_large_lr:
508 return args.legacy_large_lr
510 return args.legacy_small_lr
514 # 1) linear warmup for warmup_iter steps
515 if it < args.nb_warmup_iter:
516 return args.learning_rate * it / args.nb_warmup_iter
517 # 2) if it > nb_decay_iter, return min learning rate
518 if it > args.nb_decay_iter:
519 return args.min_learning_rate
520 # 3) in between, use cosine decay down to min learning rate
521 decay_ratio = (it - args.nb_warmup_iter) / (
522 args.nb_decay_iter - args.nb_warmup_iter
524 coeff = 0.5 * (1.0 + math.cos(math.pi * decay_ratio)) # coeff ranges 0..1
525 return args.min_learning_rate + coeff * (
526 args.learning_rate - args.min_learning_rate
530 ######################################################################
533 def add_memex_v2(batches, memex_proba, marker_token):
534 for input in batches:
535 if torch.rand(1).item() < memex_proba:
537 torch.arange(1 + 2 * input.size(1), device=input.device)[None, :]
538 .expand(input.size(0), -1)
542 u0 = torch.randint(input.size(1), (input.size(0), 1), device=input.device)
543 caterpillar_length = args.nb_lines // args.caterpillar_height
547 caterpillar_length, (input.size(0), 1), device=input.device
553 m1 = (t >= u1).long() * (t < u1 + input.size(1)).long()
555 t = t * m0 + ((-1) * (1 - m0) * (1 - m1)) + (t - u1) * m1
557 n = torch.arange(input.size(0), device=input.device)[:, None].expand(
561 new_input = input[n, t.clamp(min=0)]
562 new_input = (1 - m) * new_input + m * (marker_token)
569 def add_memex_v3(batches, memex_proba, marker_token):
570 for input in batches:
571 if torch.rand(1).item() < memex_proba:
573 torch.arange(2 * input.size(1), device=input.device)[None, :]
574 .expand(input.size(0), -1)
578 u = torch.rand(t.size(), device=t.device)
579 u[:, : input.size(1)] = 1.0
580 memex_v3_proba_fragment = 1 / 20
581 u = (u < memex_v3_proba_fragment).long()
582 v = u * torch.randint(input.size(1), u.size())
583 u[:, input.size(1) + 1 :] = v[:, input.size(1) + 1 :] - u[
584 :, : input.size(1) - 1
586 u = u.cumsum().clamp(min=0)
588 u0 = torch.randint(input.size(1), (input.size(0), 1), device=input.device)
589 caterpillar_length = args.nb_lines // args.caterpillar_height
593 caterpillar_length, (input.size(0), 1), device=input.device
599 m1 = (t >= u1).long() * (t < u1 + input.size(1)).long()
601 t = t * m0 + ((-1) * (1 - m0) * (1 - m1)) + (t - u1) * m1
603 n = torch.arange(input.size(0), device=input.device)[:, None].expand(
607 new_input = input[n, t.clamp(min=0)]
608 new_input = (1 - m) * new_input + m * (marker_token)
615 ######################################################################
617 assert args.picocvlr_prune_properties in {"none", "train+eval", "eval"}
620 def picoclvr_pruner_horizontal_green(p):
621 return not ("green" in p and ("left" in p or "right" in p))
624 picoclvr_pruner_train = (
625 picoclvr_pruner_horizontal_green
626 if args.picocvlr_prune_properties in {"train+eval"}
630 picoclvr_pruner_eval = (
631 (lambda p: not picoclvr_pruner_horizontal_green(p))
632 if args.picocvlr_prune_properties in {"train+eval", "eval"}
636 ######################################################################
640 if args.task == "byheart":
641 task = tasks.SandBox(
642 problem=problems.ProblemByHeart(),
643 nb_train_samples=args.nb_train_samples,
644 nb_test_samples=args.nb_test_samples,
645 batch_size=args.physical_batch_size,
649 args.max_percents_of_test_in_train = -1
651 elif args.task == "learnop":
652 task = tasks.SandBox(
653 problem=problems.ProblemLearnOperator(),
654 nb_train_samples=args.nb_train_samples,
655 nb_test_samples=args.nb_test_samples,
656 batch_size=args.physical_batch_size,
662 elif args.task == "guessop":
663 task = tasks.SandBox(
664 problem=problems.ProblemGuessOperator(),
665 nb_train_samples=args.nb_train_samples,
666 nb_test_samples=args.nb_test_samples,
667 batch_size=args.physical_batch_size,
673 elif args.task == "twotargets":
674 task = tasks.SandBox(
675 problem=problems.ProblemTwoTargets(),
676 nb_train_samples=args.nb_train_samples,
677 nb_test_samples=args.nb_test_samples,
678 batch_size=args.physical_batch_size,
683 elif args.task == "memory":
684 task = tasks.SandBox(
685 problem=problems.ProblemMemory(len_total=args.memory_len_total),
686 nb_train_samples=args.nb_train_samples,
687 nb_test_samples=args.nb_test_samples,
688 batch_size=args.physical_batch_size,
693 elif args.task == "mixing":
694 task = tasks.SandBox(
695 problem=problems.ProblemMixing(
696 hard=args.mixing_hard, random_start=not args.mixing_deterministic_start
698 nb_train_samples=args.nb_train_samples,
699 nb_test_samples=args.nb_test_samples,
700 batch_size=args.physical_batch_size,
705 elif args.task == "addition":
706 task = tasks.SandBox(
707 problem=problems.ProblemAddition(),
708 nb_train_samples=args.nb_train_samples,
709 nb_test_samples=args.nb_test_samples,
710 batch_size=args.physical_batch_size,
715 elif args.task == "picoclvr":
716 task = tasks.PicoCLVR(
717 nb_train_samples=args.nb_train_samples,
718 nb_test_samples=args.nb_test_samples,
719 batch_size=args.physical_batch_size,
720 height=args.picoclvr_height,
721 width=args.picoclvr_width,
722 nb_colors=args.picoclvr_nb_colors,
725 pruner_train=picoclvr_pruner_train,
726 pruner_eval=picoclvr_pruner_eval,
729 elif args.task == "mnist":
731 nb_train_samples=args.nb_train_samples,
732 nb_test_samples=args.nb_test_samples,
733 batch_size=args.physical_batch_size,
737 elif args.task == "maze":
739 nb_train_samples=args.nb_train_samples,
740 nb_test_samples=args.nb_test_samples,
741 batch_size=args.physical_batch_size,
742 height=args.maze_height,
743 width=args.maze_width,
744 nb_walls=args.maze_nb_walls,
748 elif args.task == "snake":
750 nb_train_samples=args.nb_train_samples,
751 nb_test_samples=args.nb_test_samples,
752 batch_size=args.physical_batch_size,
753 height=args.snake_height,
754 width=args.snake_width,
755 nb_colors=args.snake_nb_colors,
756 length=args.snake_length,
757 prompt_length=args.snake_length // 2,
761 elif args.task == "stack":
763 nb_train_samples=args.nb_train_samples,
764 nb_test_samples=args.nb_test_samples,
765 batch_size=args.physical_batch_size,
767 nb_steps=args.stack_nb_steps,
768 nb_stacks=args.stack_nb_stacks,
769 nb_digits=args.stack_nb_digits,
770 fraction_values_for_train=args.stack_fraction_values_for_train,
774 elif args.task == "expr":
776 nb_train_samples=args.nb_train_samples,
777 nb_test_samples=args.nb_test_samples,
778 nb_variables=args.expr_nb_variables,
779 sequence_length=args.expr_sequence_length,
780 operand_max=args.expr_operand_max,
781 result_max=args.expr_result_max,
782 batch_size=args.physical_batch_size,
786 elif args.task == "rpl":
788 nb_train_samples=args.nb_train_samples,
789 nb_test_samples=args.nb_test_samples,
790 batch_size=args.physical_batch_size,
791 nb_starting_values=args.rpl_nb_starting_values,
792 max_input=args.rpl_max_input,
793 prog_len=args.rpl_prog_len,
794 nb_runs=args.rpl_nb_runs,
795 no_prog=args.rpl_no_prog,
800 elif args.task == "grid":
802 nb_train_samples=args.nb_train_samples,
803 nb_test_samples=args.nb_test_samples,
804 batch_size=args.physical_batch_size,
806 nb_shapes=args.grid_nb_shapes,
807 nb_colors=args.grid_nb_colors,
812 elif args.task == "qmlp":
814 nb_train_samples=args.nb_train_samples,
815 nb_test_samples=args.nb_test_samples,
816 batch_size=args.physical_batch_size,
817 result_dir=args.result_dir,
823 raise ValueError(f"Unknown task {args.task}")
825 ######################################################################
827 log_string(f"device {device}")
829 vocabulary_size = task.vocabulary_size()
831 if args.memex_proba > 0:
834 log_string(f"vocabulary_size {vocabulary_size}")
836 ##############################
839 vocabulary_size=vocabulary_size,
840 dim_model=args.dim_model,
841 dim_keys=args.dim_keys,
842 dim_hidden=args.dim_hidden,
843 nb_heads=args.nb_heads,
844 nb_lines=args.nb_lines,
845 caterpillar_height=args.caterpillar_height,
846 nb_blocks=args.nb_blocks,
848 dropout=args.dropout,
849 attention_layer=args.attention,
856 nb_parameters = sum(p.numel() for p in model.parameters())
857 log_string(f"nb_parameters {nb_parameters} ({int(nb_parameters/1e6)}M)")
859 ######################################################################
861 nb_epochs_finished = 0
863 if args.no_checkpoint:
864 log_string(f"not trying to load checkpoint.")
868 checkpoint_name = os.path.join(args.result_dir, args.checkpoint_name)
869 checkpoint = torch.load(checkpoint_name)
870 nb_epochs_finished = checkpoint["nb_epochs_finished"]
871 model.load_state_dict(checkpoint["model_state"])
872 torch.set_rng_state(checkpoint["rng_state"])
873 if torch.cuda.is_available():
874 torch.cuda.set_rng_state(checkpoint["cuda_rng_state"])
876 log_string(f"checkpoint loaded with {nb_epochs_finished} epochs finished.")
878 except FileNotFoundError:
879 log_string("starting from scratch.")
882 log_string("error when loading the checkpoint.")
885 ######################################################################
887 if args.task == "expr" and args.expr_input_file is not None:
888 task.produce_results(
889 n_epoch=nb_epochs_finished,
891 result_dir=args.result_dir,
893 deterministic_synthesis=args.deterministic_synthesis,
894 input_file=args.expr_input_file,
899 ######################################################################
901 nb_epochs = args.nb_epochs if args.nb_epochs > 0 else nb_epochs_default
903 # Compute the entropy of the training tokens
906 for input in task.batches(split="train"):
907 token_count += F.one_hot(input, num_classes=task.vocabulary_size()).sum((0, 1))
908 token_probas = token_count / token_count.sum()
909 entropy = -torch.xlogy(token_probas, token_probas).sum()
910 train_set_perplexity = math.exp(entropy)
912 ######################################################################
913 # A bit of paranoia never hurts
915 if args.max_percents_of_test_in_train >= 0:
917 def subsets_as_tuples(batches, cs):
919 for batch in batches:
921 s.add(tuple([v.item() for v in x]))
927 nb_test, nb_in_train = 0, 0
928 for test_subset in subsets_as_tuples(task.batches(split="test"), 25000):
930 for train_subset in subsets_as_tuples(task.batches(split="train"), 25000):
931 in_train.update(test_subset.intersection(train_subset))
932 nb_in_train += len(in_train)
933 nb_test += len(test_subset)
936 f"data_check {nb_in_train*100/nb_test:.02f}% ({nb_in_train}/{nb_test}) of test samples are in the train set"
940 nb_in_train <= args.max_percents_of_test_in_train * nb_test / 100
941 ), f"More than {args.max_percents_of_test_in_train}% of test samples are in the train set"
943 ##############################
945 if "calibrate" in sup_args:
946 for input in task.batches(split="train", desc="calibrate"):
947 input = input.to(device)
948 output = model(mygpt.BracketedSequence(input)).x
950 for n, m in model.named_modules():
953 if isinstance(x, mygpt.Calibrator):
954 print(f"####### ${n} | ${a} ########################")
955 mean, std = x.moments()
956 print("mean\n", mean, "\n")
957 print("std\n", std, "\n")
958 print(f"############################################\n\n")
962 ##############################
966 if nb_epochs_finished >= nb_epochs:
967 task.produce_results(
968 n_epoch=nb_epochs_finished,
970 result_dir=args.result_dir,
972 deterministic_synthesis=args.deterministic_synthesis,
975 time_pred_result = datetime.datetime.now()
981 for n_epoch in range(nb_epochs_finished, nb_epochs):
982 if args.optim == "sgd":
983 optimizer = torch.optim.SGD(model.parameters(), lr=args.learning_rate)
984 elif args.optim == "adam":
985 optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate)
986 elif args.optim == "adamw":
987 optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate)
989 raise ValueError(f"Unknown optimizer {args.optim}.")
993 nb_train_samples, acc_train_loss, acc_train_inner_loss = 0, 0.0, 0.0
997 if args.memex_nb_epochs is None or n_epoch < args.memex_nb_epochs
1001 log_string(f"memex_proba {memex_proba}")
1003 train_batches = add_memex_v2(
1004 batches=task.batches(split="train"),
1005 memex_proba=memex_proba,
1006 marker_token=vocabulary_size - 1,
1016 for input in add_none(train_batches):
1017 if input is not None:
1018 model.reset_inner_loss()
1019 input = input.to(device)
1021 output = model(mygpt.BracketedSequence(input)).x
1022 loss = F.cross_entropy(output.transpose(1, 2), input)
1023 inner_loss = model.get_inner_loss()
1025 acc_train_loss += loss.item() * input.size(0)
1026 acc_train_inner_loss += inner_loss.item() * input.size(0)
1028 nb_train_samples += input.size(0)
1029 nb_samples_seen += input.size(0)
1031 total_loss = loss + (
1032 args.rho_inner_loss * inner_loss if args.rho_inner_loss > 0 else 0.0
1036 lr = get_lr(n_epoch, it)
1037 for param_group in optimizer.param_groups:
1038 param_group["lr"] = lr
1040 # log_string(f"learning_rate {lr}")
1042 total_loss.backward()
1043 nb_acc_samples += input.size(0)
1045 if (input is None and nb_acc_samples > 0) or nb_acc_samples == args.batch_size:
1046 assert nb_acc_samples <= args.batch_size
1048 grad_norm = sum([p.grad.pow(2).sum() for p in model.parameters()]).sqrt()
1049 loss_file.write(f"{n_epoch} {n_batch} {loss.item()} {grad_norm.item()}\n")
1050 optimizer.zero_grad()
1055 with torch.autograd.no_grad():
1058 nb_test_samples, acc_test_loss = 0, 0.0
1060 for input in task.batches(split="test"):
1061 input = input.to(device)
1063 output = model(mygpt.BracketedSequence(input)).x
1064 loss = F.cross_entropy(output.transpose(1, 2), input)
1065 acc_test_loss += loss.item() * input.size(0)
1066 nb_test_samples += input.size(0)
1069 f"loss {n_epoch} train_loss {acc_train_loss/nb_train_samples} train_inner_loss {acc_train_inner_loss/nb_train_samples} test_prediction {acc_test_loss/nb_test_samples}"
1072 task.produce_results(
1075 result_dir=args.result_dir,
1077 deterministic_synthesis=args.deterministic_synthesis,
1080 train_perplexity = math.exp(min(100, acc_train_loss / nb_train_samples))
1081 test_perplexity = math.exp(min(100, acc_test_loss / nb_test_samples))
1084 f"perplexity {n_epoch} train_set {train_set_perplexity} train_prediction {train_perplexity} test_prediction {test_perplexity}"
1087 time_current_result = datetime.datetime.now()
1089 f"next_result {time_current_result + (time_current_result - time_pred_result)}"
1091 time_pred_result = time_current_result
1094 "nb_epochs_finished": n_epoch + 1,
1095 "model_state": model.state_dict(),
1096 "rng_state": torch.get_rng_state(),
1099 if torch.cuda.is_available():
1100 checkpoint["cuda_rng_state"] = torch.cuda.get_rng_state()
1102 checkpoint_name = os.path.join(args.result_dir, args.checkpoint_name)
1103 torch.save(checkpoint, checkpoint_name)
1104 log_string(f"saved checkpoint {checkpoint_name}")
1106 ######################################################################