fix random process

reproduction/Star_transformer/train.py

@@ -1,4 +1,6 @@
 from util import get_argparser, set_gpu, set_rng_seeds, add_model_args
+seed = set_rng_seeds(15360)
+print('RNG SEED {}'.format(seed))
 from datasets import load_seqtag, load_sst, load_snli, EmbedLoader, MAX_LEN
 import torch.nn as nn
 import torch
@@ -109,9 +111,6 @@ class MyCallback(FN.core.callback.Callback):
 
 
 def train():
-    seed = set_rng_seeds(28848)
-    #seed = set_rng_seeds(np.random.randint(65536))
-    print('RNG SEED {}'.format(seed))
     print('loading data')
     ds_list, word_v, tag_v = g_datasets['{}-{}'.format(
         g_args.ds, g_args.task)]()