X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=main.py;h=ba8843b313993062d3dd4b8f1020de77b5bf71e3;hb=d7eeacf1eab237bbbe67d3e44b90b57fd1445667;hp=58e80462609bcda7cb8bc43f88d5dfec9022b8d4;hpb=8e23dd068df00df61c690ffa89ecc8cb9db4b32d;p=picoclvr.git diff --git a/main.py b/main.py index 58e8046..ba8843b 100755 --- a/main.py +++ b/main.py @@ -5,15 +5,13 @@ # Written by Francois Fleuret -# torch.backends.cuda.matmul.allow_tf23 -# torch.autocast(torch.bfloat16) - import math, sys, argparse, time, tqdm, os import torch, torchvision from torch import nn from torch.nn import functional as F +import ffutils import mygpt, tasks ###################################################################### @@ -34,8 +32,8 @@ 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 +56,17 @@ 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_model", type=int, default=None) -parser.add_argument("--dim_keys", type=int, default=64) +parser.add_argument("--dim_keys", type=int, default=None) -parser.add_argument("--dim_hidden", type=int, default=2048) +parser.add_argument("--dim_hidden", type=int, default=None) -parser.add_argument("--nb_heads", type=int, default=8) +parser.add_argument("--nb_heads", type=int, default=None) -parser.add_argument("--nb_blocks", type=int, default=12) +parser.add_argument("--nb_blocks", type=int, default=None) parser.add_argument("--dropout", type=float, default=0.1) @@ -78,6 +78,28 @@ parser.add_argument("--overwrite_results", action="store_true", default=False) parser.add_argument("--checkpoint_name", type=str, default="checkpoint.pth") +############################## +# rpl options + +parser.add_argument("--rpl-nb_starting_values", type=int, default=5) + +parser.add_argument("--rpl-max_input", type=int, default=9) + +parser.add_argument("--rpl-prog_len", type=int, default=10) + +parser.add_argument("--rpl-nb_runs", type=int, default=8) + +############################## +# sandbox 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 @@ -136,7 +158,7 @@ parser.add_argument("--expr_input_file", type=str, default=None) ############################## # World options -parser.add_argument("--world_vqae_nb_epochs", type=int, default=10) +parser.add_argument("--world_vqae_nb_epochs", type=int, default=25) ###################################################################### @@ -149,7 +171,13 @@ 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, @@ -186,18 +214,64 @@ 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}") ###################################################################### @@ -257,7 +331,38 @@ 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, @@ -328,12 +433,26 @@ 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, + nb_starting_values=args.rpl_nb_starting_values, + max_input=args.rpl_max_input, + prog_len=args.rpl_prog_len, + nb_runs=args.rpl_nb_runs, + 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, )