Update train.py

Fix the tokenizer learning

train.py

@@ -10,7 +10,7 @@ with open("dataset.json", "r") as f:
     dset = json.load(f)
 
 tokenizer = Tokenizer()
-tokenizer.fit_on_texts(dset)
+tokenizer.fit_on_texts(list(dset.keys()))
 
 emb_size = 128 # how big are the word vectors in the input (how much information can be fit into one word)
 vocab_size = len(tokenizer.get_vocabulary())