diff --git a/model.py b/model.py
index 4ee7cfc1..93331ad9 100644
--- a/model.py
+++ b/model.py
@@ -3,6 +3,7 @@
 from tensorflow.python.ops import seq2seq
 
 import numpy as np
+import itertools
 
 class Model():
     def __init__(self, args, infer=False):
@@ -58,7 +59,7 @@ def loop(prev, _):
         optimizer = tf.train.AdamOptimizer(self.lr)
         self.train_op = optimizer.apply_gradients(zip(grads, tvars))
 
-    def sample(self, sess, chars, vocab, num=200, prime='The ', sampling_type=1):
+    def stream(self, sess, chars, vocab, prime=u'The ', sampling_type=1):
         state = self.cell.zero_state(1, tf.float32).eval()
         for char in prime[:-1]:
             x = np.zeros((1, 1))
@@ -71,9 +72,11 @@ def weighted_pick(weights):
             s = np.sum(weights)
             return(int(np.searchsorted(t, np.random.rand(1)*s)))
 
-        ret = prime
+        for char in prime:
+            yield char
+
         char = prime[-1]
-        for n in range(num):
+        while True:
             x = np.zeros((1, 1))
             x[0, 0] = vocab[char]
             feed = {self.input_data: x, self.initial_state:state}
@@ -90,9 +93,10 @@ def weighted_pick(weights):
             else: # sampling_type == 1 default:
                 sample = weighted_pick(p)
 
-            pred = chars[sample]
-            ret += pred
-            char = pred
-        return ret
+            char = chars[sample]
+            yield char
 
+    def sample(self, sess, chars, vocab, num=200, prime=u'The ', sampling_type=1):
+        stream = self.stream(sess, chars, vocab, prime=prime, sampling_type=sampling_type)
+        return u''.join(itertools.islice(stream, num))
 
diff --git a/sample.py b/sample.py
index d3f04285..56f5a3a4 100644
--- a/sample.py
+++ b/sample.py
@@ -9,6 +9,7 @@
 
 from utils import TextLoader
 from model import Model
+import sys
 
 from six import text_type
 
@@ -17,7 +18,7 @@ def main():
     parser.add_argument('--save_dir', type=str, default='save',
                        help='model directory to store checkpointed models')
     parser.add_argument('-n', type=int, default=500,
-                       help='number of characters to sample')
+                       help='number of characters to sample, -1 to sample forever')
     parser.add_argument('--prime', type=text_type, default=u' ',
                        help='prime text')
     parser.add_argument('--sample', type=int, default=1,
@@ -38,7 +39,11 @@ def sample(args):
         ckpt = tf.train.get_checkpoint_state(args.save_dir)
         if ckpt and ckpt.model_checkpoint_path:
             saver.restore(sess, ckpt.model_checkpoint_path)
-            print(model.sample(sess, chars, vocab, args.n, args.prime, args.sample))
+            if args.n > 0:
+                print(model.sample(sess, chars, vocab, args.n, args.prime, args.sample))
+            else:
+                for char in model.stream(sess, chars, vocab, args.prime, args.sample):
+                    sys.stdout.write(char)
 
 if __name__ == '__main__':
     main()