update

serveur/gpt-2/src/encoder.py

@@ -116,3 +116,4 @@ def get_encoder(model_name):
         encoder=encoder,
         bpe_merges=bpe_merges,
     )
+

serveur/gpt-2/src/generate_unconditional_samples.py

@@ -41,7 +41,7 @@ def sample_model(
     """
     enc = encoder.get_encoder(model_name)
     hparams = model.default_hparams()
-    with open(os.path.join('models', model_name, 'hparams.json')) as f:
+    with open(os.path.join('../models', model_name, 'hparams.json')) as f:
         hparams.override_from_dict(json.load(f))
 
     if length is None:
@@ -61,7 +61,7 @@ def sample_model(
         )[:, 1:]
 
         saver = tf.train.Saver()
-        ckpt = tf.train.latest_checkpoint(os.path.join('models', model_name))
+        ckpt = tf.train.latest_checkpoint(os.path.join('../models', model_name))
         saver.restore(sess, ckpt)
 
         generated = 0
@@ -71,7 +71,7 @@ def sample_model(
                 generated += batch_size
                 text = enc.decode(out[i])
                 print("=" * 40 + " SAMPLE " + str(generated) + " " + "=" * 40)
-                path1 = "/home/stu/pkq/gpt-2/samples/new_try_samples/uncon"
+                path1 = '../samples/story_unconditional'
                 path2 = str(generated) + ".txt"
                 path = os.path.join(path1, path2)
 
@@ -82,4 +82,4 @@ def sample_model(
         return 1
 
 if __name__ == '__main__':
-    fire.Fire(sample_model)
+    sample_model(model_name="345MShort", nsamples=3)

serveur/gpt-2/src/run.py

@@ -18,5 +18,5 @@ def run1(path1):
 
 
 #run1("/home/stu/pkq/gpt-2/samples/story_start/1_start.txt")
-raw_text = ""
+raw_text = " "
 model2.train(raw_text, 5)
\ No newline at end of file