Update run.py

run.py

@@ -16,7 +16,7 @@ def response(message, history, model):
   params={"max_length":500, "return_full_text":False} #, "stream":True
   url = f"https://api-inference.huggingface.co/models/{model_id}"
   correction=1
-  prompt=f"<s>[INST] {message} [/INST]" 
+  prompt=f"[INST] {message} [/INST]" # skipped <s>
   print("URL: "+url)
   print(params)
   print("User: "+message+"\nAI: ")
@@ -34,7 +34,7 @@ def response(message, history, model):
       text=text.rstrip('"}]')
       correction=3 
     response=response+text
-    print(response)
+    print(text)
     time.sleep(0.2)
     yield response