######################################################################
-args = parser.parse_args()
+# args = parser.parse_args()
-assert args.picocvlr_prune_properties in {"none", "train+eval", "eval"}
+args, sup_args = parser.parse_known_args()
+
+sup_args = dict([x.removeprefix("--").split("=") for x in sup_args])
if args.result_dir is None:
args.result_dir = f"results_{args.task}_{args.model}"
print(f"result directory {args.result_dir} already exists")
exit(1)
+loss_file = open(os.path.join(args.result_dir, "loss.dat"), "a")
+
log_file = open(os.path.join(args.result_dir, args.log_filename), "a")
if args.seed >= 0:
for n in vars(args):
log_string(f"args.{n} {getattr(args, n)}")
+for k, v in sup_args.items():
+ log_string(f'sup_args["{k}"] "{v}"')
+
######################################################################
######################################################################
+assert args.picocvlr_prune_properties in {"none", "train+eval", "eval"}
+
+
def picoclvr_pruner_horizontal_green(p):
return not ("green" in p and ("left" in p or "right" in p))
causal=True,
dropout=args.dropout,
attention_layer=args.attention,
+ logger=log_string,
+ **sup_args,
)
model.to(device)
it = 0
+n_batch = 0
+
for n_epoch in range(nb_epochs_finished, nb_epochs):
if args.optim == "sgd":
optimizer = torch.optim.SGD(model.parameters(), lr=args.learning_rate)
total_loss.backward()
optimizer.step()
+ grad_norm = sum([p.grad.pow(2).sum() for p in model.parameters()]).sqrt()
+
+ loss_file.write(f"{n_epoch} {n_batch} {loss.item()} {grad_norm.item()}\n")
+
+ n_batch += 1
+
with torch.autograd.no_grad():
model.eval()