X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=main.py;h=901b1d0529bb525b1cbca2b5c3bc91af7b12bf36;hb=439c597d409c344283f8996f042daf79d3f24de2;hp=c763016bb7e06be9252dda4996b2f74c053732c8;hpb=bf48dc69f7f57ad391481c8917570e35f661cc4a;p=picoclvr.git diff --git a/main.py b/main.py index c763016..901b1d0 100755 --- a/main.py +++ b/main.py @@ -14,6 +14,7 @@ import torch, torchvision from torch import nn from torch.nn import functional as F +import ffutils import mygpt, tasks ###################################################################### @@ -34,8 +35,8 @@ parser = argparse.ArgumentParser( parser.add_argument( "--task", type=str, - default="picoclvr", - help="picoclvr, mnist, maze, snake, stack, expr, world", + default="sandbox", + help="sandbox, picoclvr, mnist, maze, snake, stack, expr, rpl, world", ) parser.add_argument("--log_filename", type=str, default="train.log", help=" ") @@ -58,15 +59,17 @@ parser.add_argument("--learning_rate", type=float, default=1e-4) parser.add_argument("--learning_rate_schedule", type=str, default="10: 2e-5,30: 4e-6") -parser.add_argument("--dim_model", type=int, default=512) +parser.add_argument("--model", type=str, default="37M") -parser.add_argument("--dim_keys", type=int, default=64) +parser.add_argument("--dim_model", type=int, default=None) -parser.add_argument("--dim_hidden", type=int, default=2048) +parser.add_argument("--dim_keys", type=int, default=None) -parser.add_argument("--nb_heads", type=int, default=8) +parser.add_argument("--dim_hidden", type=int, default=None) -parser.add_argument("--nb_blocks", type=int, default=12) +parser.add_argument("--nb_heads", type=int, default=None) + +parser.add_argument("--nb_blocks", type=int, default=None) parser.add_argument("--dropout", type=float, default=0.1) @@ -81,6 +84,17 @@ parser.add_argument("--checkpoint_name", type=str, default="checkpoint.pth") ############################## # picoclvr options +parser.add_argument("--sandbox_level", type=int, default=0) + +parser.add_argument("--sandbox_levels_nb_items", type=int, default=25) + +parser.add_argument("--sandbox_levels_len_source", type=int, default=6) + +parser.add_argument("--sandbox_levels_len_result", type=int, default=8) + +############################## +# picoclvr options + parser.add_argument("--picoclvr_nb_colors", type=int, default=5) parser.add_argument("--picoclvr_height", type=int, default=12) @@ -133,6 +147,11 @@ parser.add_argument("--expr_result_max", type=int, default=99) parser.add_argument("--expr_input_file", type=str, default=None) +############################## +# World options + +parser.add_argument("--world_vqae_nb_epochs", type=int, default=25) + ###################################################################### args = parser.parse_args() @@ -144,7 +163,13 @@ if args.result_dir is None: ###################################################################### -default_args = { +default_task_args = { + "sandbox": { + "nb_epochs": 50, + "batch_size": 25, + "nb_train_samples": 100000, + "nb_test_samples": 10000, + }, "picoclvr": { "nb_epochs": 25, "batch_size": 25, @@ -181,21 +206,67 @@ default_args = { "nb_train_samples": 1000000, "nb_test_samples": 10000, }, + "rpl": { + "nb_epochs": 40, + "batch_size": 25, + "nb_train_samples": 100000, + "nb_test_samples": 10000, + }, "world": { - "nb_epochs": 5, + "nb_epochs": 10, "batch_size": 25, - "nb_train_samples": 10000, + "nb_train_samples": 25000, "nb_test_samples": 1000, }, } -if args.task in default_args: - for k, v in default_args[args.task].items(): +if args.task in default_task_args: + for k, v in default_task_args[args.task].items(): if getattr(args, k) is None: setattr(args, k, v) ###################################################################### +default_model_args = { + "17K": { + "dim_model": 32, + "dim_keys": 32, + "dim_hidden": 32, + "nb_heads": 2, + "nb_blocks": 2, + }, + "37M": { + "dim_model": 512, + "dim_keys": 64, + "dim_hidden": 2048, + "nb_heads": 8, + "nb_blocks": 12, + }, + "122M": { + "dim_model": 768, + "dim_keys": 64, + "dim_hidden": 2048, + "nb_heads": 8, + "nb_blocks": 24, + }, + "352M": { + "dim_model": 1024, + "dim_keys": 64, + "dim_hidden": 2048, + "nb_heads": 8, + "nb_blocks": 48, + }, +} + +if args.model in default_model_args: + for k, v in default_model_args[args.model].items(): + if getattr(args, k) is None: + setattr(args, k, v) +else: + raise ValueError(f"Unknown model {args.model}") + +###################################################################### + try: os.mkdir(args.result_dir) except FileExistsError: @@ -252,7 +323,38 @@ picoclvr_pruner_eval = ( ###################################################################### -if args.task == "picoclvr": +if args.task == "sandbox": + if args.sandbox_level == 0: + problem = tasks.ProblemLevel0( + nb_sentences=args.sandbox_levels_nb_items, + len_prompt=args.sandbox_levels_len_source, + len_result=args.sandbox_levels_len_result, + ) + elif args.sandbox_level == 1: + problem = tasks.ProblemLevel1( + nb_operators=args.sandbox_levels_nb_items, + len_source=args.sandbox_levels_len_source, + len_result=args.sandbox_levels_len_result, + ) + elif args.sandbox_level == 2: + problem = tasks.ProblemLevel2( + len_source=args.sandbox_levels_len_source, + len_result=args.sandbox_levels_len_result, + ) + else: + raise ValueError(f"Unknown sandbox level {args.sandbox_level}") + + task = tasks.SandBox( + problem, + # tasks.ProblemAddition(zero_padded=False, inverted_result=False), + nb_train_samples=args.nb_train_samples, + nb_test_samples=args.nb_test_samples, + batch_size=args.batch_size, + logger=log_string, + device=device, + ) + +elif args.task == "picoclvr": task = tasks.PicoCLVR( nb_train_samples=args.nb_train_samples, nb_test_samples=args.nb_test_samples, @@ -323,11 +425,22 @@ elif args.task == "expr": device=device, ) +elif args.task == "rpl": + task = tasks.RPL( + nb_train_samples=args.nb_train_samples, + nb_test_samples=args.nb_test_samples, + batch_size=args.batch_size, + logger=log_string, + device=device, + ) + elif args.task == "world": task = tasks.World( nb_train_samples=args.nb_train_samples, nb_test_samples=args.nb_test_samples, batch_size=args.batch_size, + vqae_nb_epochs=args.world_vqae_nb_epochs, + logger=log_string, device=device, )