X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=main.py;h=901b1d0529bb525b1cbca2b5c3bc91af7b12bf36;hb=439c597d409c344283f8996f042daf79d3f24de2;hp=0d4930dc32104f7dd3bf579262f4ab7e82746d03;hpb=d6f73f1d5093fb098e822e14db382dd3a1c63a2a;p=picoclvr.git

diff --git a/main.py b/main.py
index 0d4930d..901b1d0 100755
--- a/main.py
+++ b/main.py
@@ -36,7 +36,7 @@ parser.add_argument(
     "--task",
     type=str,
     default="sandbox",
-    help="sandbox, picoclvr, mnist, maze, snake, stack, expr, world",
+    help="sandbox, picoclvr, mnist, maze, snake, stack, expr, rpl, world",
 )
 
 parser.add_argument("--log_filename", type=str, default="train.log", help=" ")
@@ -59,15 +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)
 
@@ -86,7 +88,7 @@ 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=5)
+parser.add_argument("--sandbox_levels_len_source", type=int, default=6)
 
 parser.add_argument("--sandbox_levels_len_result", type=int, default=8)
 
@@ -161,11 +163,11 @@ if args.result_dir is None:
 
 ######################################################################
 
-default_args = {
+default_task_args = {
     "sandbox": {
-        "nb_epochs": 10,
+        "nb_epochs": 50,
         "batch_size": 25,
-        "nb_train_samples": 25000,
+        "nb_train_samples": 100000,
         "nb_test_samples": 10000,
     },
     "picoclvr": {
@@ -204,6 +206,12 @@ 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": 10,
         "batch_size": 25,
@@ -212,10 +220,50 @@ default_args = {
     },
 }
 
-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}")
 
 ######################################################################
 
@@ -377,6 +425,15 @@ 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,