Small fixes.

hypertrain.py

@@ -214,7 +214,7 @@ parser.add_argument('--optimizer', type=str, help='the optimizer to use',
 parser.add_argument('--lr', type=float, help='learning rate', default=2e-4)
 parser.add_argument('--end_lr', type=float, help='end learning rate',
                     default=2e-4)
-parser.add_argument('--warmup', type=int, help='warmup steps')
+parser.add_argument('--warmup', type=int, help='warmup steps', default=10)
 parser.add_argument('--bs', type=int, help='batch size', default=4)
 parser.add_argument('--gas', type=int, help='gas', default=1)
 parser.add_argument('--seed', type=int, help="Random seed value",
@@ -247,7 +247,7 @@ if args.output == '':
 # we need 250 batch size to train the small GPT.
 train_config = {
     "data_path": args.dataset,
-    "save_path": args.model,
+    "save_path": args.output,
     "lm_path": args.model,
     "optimizer": args.optimizer,
     "masked_softmax_fusion": args.masked,
@@ -259,7 +259,7 @@ train_config = {
     "bs": args.bs,
     "gas": args.gas,
     "seed": args.seed,
-    "save_every": args.save_steps0,
+    "save_every": args.save_steps,
     "amp": args.amp,
     "loss_scale": args.loss_scale,
     "eval_every": args.eval_every,