Update app.py

Change search for next word from greedy to beam.

app.py

@@ -171,7 +171,14 @@ gpt2_tkn=GPT2Tokenizer.from_pretrained('gpt2')
 
 def generate(starting_text):
     tkn_ids = gpt2_tkn.encode(starting_text, return_tensors = 'pt')
-    gpt2_tensors = mdl.generate(tkn_ids, max_length=100, no_repeat_ngram_size=True)
+    
+    # When no specific parameter is specified, the model performs a greedy search to find the next word, which entails selecting the word from all of the
+    # alternatives that has the highest probability of being correct. This process is deterministic in nature, which means that resultant text is the same
+    # as before if we use the same parameters.
+
+    # The num_beams parameter does a beam search: it returns the sequences that have the highest probability, and then, when it comes time to
+    # choose, it picks the one that has the highest probability.
+    gpt2_tensors = mdl.generate(tkn_ids, max_length=100, no_repeat_ngram_size=True, num_beams=3)
     response=""
     #response = gpt2_tensors
     for i, x in enumerate(gpt2_tensors):