"nb_train_samples": 250000,
"nb_test_samples": 10000,
},
+ "world": {
+ "model": "37M",
+ "batch_size": 25,
+ "nb_train_samples": 50000,
+ "nb_test_samples": 10000,
+ },
"byheart": {
"model": "37M",
"batch_size": 25,
)
args.max_percents_of_test_in_train = -1
+elif args.task == "world":
+ task = tasks.World(
+ nb_train_samples=args.nb_train_samples,
+ nb_test_samples=args.nb_test_samples,
+ batch_size=args.physical_batch_size,
+ logger=log_string,
+ device=device,
+ )
+ args.max_percents_of_test_in_train = -1
+
elif args.task == "learnop":
task = tasks.SandBox(
problem=problems.ProblemLearnOperator(),
input = input.to(device)
bs = model(mygpt.BracketedSequence(input))
- output_ar = bs.x
+ output = bs.x
loss = F.cross_entropy(output.transpose(1, 2), input)