Update app.py

app.py

@@ -25,7 +25,7 @@ def translate(input, output, text):
             "early_stopping": False,
             "length_penalty": 0.0,
             "eos_token_id": None,
-            "max_new_tokens": 2000,
+            "max_new_tokens": 250,
             "top_p": 0.9,
         }, 
         "options": {
@@ -34,9 +34,9 @@ def translate(input, output, text):
             },
     }
     response = requests.request("POST", API_URL,  json=json_) 
-    output = response.json()
+    output = response.json()[0]['generated_text']
 
-    return output
+    return output.replace(instruction, '', 1)
 
 demo = gr.Blocks()