Update app.py

app.py

@@ -22,8 +22,7 @@ def query(payload):
     return json.loads(response.content.decode("utf-8"))
 
 
-def inference(input_sentence, max_length, sample_or_greedy, seed=42):
-    sample_or_greedy = "Greedy"
+def inference(input_sentence, max_length, sample_or_greedy="Greedy", seed=42):
     if sample_or_greedy == "Sample":
         parameters = {
             "max_new_tokens": max_length,